OpenAI Releases Open-Weight AI Models, Marking Shift in Strategy

OpenAI launches two 'open' AI reasoning models | TechCrunch

OpenAI announced Tuesday the launch of two open-weight AI reasoning models with similar capabilities to its o-series. Both are freely available to download from the online developer platform, Hugging Face, the company said, describing the models as "state-of-the-art" when measured across several benchmarks for comparing open models. The models come in two sizes: a larger and more capable gpt-oss-120b model that can run on a single Nvidia GPU, and a lighter-weight gpt-oss-20b model that can run on a consumer laptop with 16GB of memory. The launch marks OpenAI's first 'open' language model since GPT-2, which was released more than five years ago. In a briefing, OpenAI said its open models will be capable of sending complex queries to AI models in the cloud, as TechCrunch previously reported. That means if OpenAI's open model is not capable of a certain task, such as processing an image, developers can connect the open model to one of the company's more capable closed models. While OpenAI open-sourced AI models in its early days, the company has generally favored a proprietary, closed-source development approach. The latter strategy has helped OpenAI build a large business selling access to its AI models via an API to enterprises and developers. However, CEO Sam Altman said in January he believes OpenAI has been "on the wrong side of history" when it comes to open sourcing its technologies. The company today faces growing pressure from Chinese AI labs -- including DeepSeek, Alibaba's Qwen, and Moonshot AI -- which have developed several of the world's most capable and popular open models. (While Meta previously dominated the open AI space, the company's Llama AI models have fallen behind in the last year.) In July, the Trump Administration also urged U.S. AI developers to open source more technology to promote global adoption of AI aligned with American values. With the release of gpt-oss, OpenAI hopes to curry favor with developers and the Trump Administration alike, both of which have watched the Chinese AI labs rise to prominence in the open source space. "Going back to when we started in 2015, OpenAI's mission is to ensure AGI that benefits all of humanity," said OpenAI CEO Sam Altman in a statement shared with TechCrunch. "To that end, we are excited for the world to be building on an open AI stack created in the United States, based on democratic values, available for free to all and for wide benefit." OpenAI aimed to make its open model a leader among other open-weight AI models, and the company claims to have done just that. On Codeforces (with tools), a competitive coding test, gpt-oss-120b and gpt-oss-20b score 2622 and 2516, respectively, outperformed DeepSeek's R1 while underperforming o3 and o4-mini. On Humanity's Last Exam, a challenging test of crowd-sourced questions across a variety of subjects (with tools), gpt-oss-120b and gpt-oss-20b score 19% and 17.3%, respectively. Similarly, this underperforms o3 but outperforms leading open models from DeepSeek and Qwen. Notably, OpenAI's open models hallucinate significantly more than its latest AI reasoning models, o3 and o4-mini. Hallucinations have been getting more severe in OpenAI's latest AI reasoning models, and the company previously said it doesn't quite understand why. In a white paper, OpenAI says this is "expected, as smaller models have less world knowledge than larger frontier models and tend to hallucinate more." OpenAI found that gpt-oss-120b and gpt-oss-20b hallucinated in response to 49% and 53% of questions on PersonQA, the company's in-house benchmark for measuring the accuracy of a model's knowledge about people. That's more than triple the hallucination rate of OpenAI's o1 model, which scored 16%, and higher than its o4-mini model, which scored 36%. OpenAI says its open models were trained with similar processes to its proprietary models. The company says each open model leverages mixture-of-experts (MoE) to tap fewer parameters for any given question, making it run more efficiently. For gpt-oss-120b, which has 117 billion total parameters, OpenAI says the model only activates 5.1 billion parameters per token. The company also says its open model was trained using high-compute reinforcement learning (RL) -- a post-training process to teach AI models right from wrong in simulated environments using large clusters of Nvidia GPUs. This was also used to train OpenAI's o-series of models, and the open models have a similar chain-of-thought process in which they take additional time and computational resources to work through their answers. As a result of the post-training process, OpenAI says its open AI models excel at powering AI agents, and are capable of calling tools such as web search or Python code execution as part of its chain-of-thought process. However, OpenAI says its open models are text-only, meaning they will not be able to process or generate images and audio like the company's other models. OpenAI is releasing gpt-oss-120b and gpt-oss-20b under the Apache 2.0 license, which is generally considered one of the most permissive. This license will allow enterprises to monetize OpenAI's open models without having to pay or obtain permission from the company. However, unlike fully open source offerings from AI labs like AI2, OpenAI says it will not be releasing the training data used to create its open models. This decision is not surprising given that several active lawsuits against AI model providers, including OpenAI, have alleged that these companies inappropriately trained their AI models on copyrighted works. OpenAI delayed the release of its open models several times in recent months, partially to address safety concerns. Beyond the company's typical safety policies, OpenAI says in a white paper that it also investigated whether bad actors could fine-tune its gpt-oss models to be more helpful in cyber attacks or the creation of biological or chemical weapons. After testing from OpenAI and third-party evaluators, the company says gpt-oss may marginally increase biological capabilities. However, it did not find evidence that these open models could reach its "high capability' threshold for danger in these domains, even after fine-tuning. While OpenAI's model appears to be state-of-the-art among open models, developers are eagerly awaiting the release of DeepSeek R2, its next AI reasoning model, as well as a new open model from Meta's new superintelligence lab.

OpenAI has finally released open-weight language models

That's particularly notable at a time when Meta, which had previously dominated the American open-model landscape with its Llama models, may be reorienting toward closed releases -- and when Chinese open models, such as DeepSeek's offerings, Kimi K2, and Alibaba's Qwen series, are becoming more popular than their American competitors. "The vast majority of our [enterprise and startup] customers are already using a lot of open models," said Casey Dvorak, a research program manager at OpenAI, in a media briefing about the model release. "Because there is no [competitive] open model from OpenAI, we wanted to plug that gap and actually allow them to use our technology across the board." The new models come in two different sizes, the smaller of which can theoretically run on 16 GB of RAM -- the minimum amount that Apple currently offers on its computers. The larger model requires a high-end laptop or specialized hardware. Open models have a few key use cases. Some organizations may want to customize models for their own purposes or save money by running models on their own equipment, though that equipment comes at a substantial upfront cost. Others -- such hospitals, law firms, and governments -- might need models that they can run locally for data security reasons. OpenAI has facilitated such activity by releasing its open models under a permissive Apache 2.0 license, which allows the models to be used for commercial purposes. Nathan Lambert, post-training lead at the Allen Institute for AI, says that this choice is commendable: Such licenses are typical for Chinese open-model releases, but Meta released its Llama models under a bespoke, more restrictive license. "It's a very good thing for the open community," he says. Researchers who study how LLMs work also need open models, so that they can examine and manipulate those models in detail. "In part, this is about reasserting OpenAI's dominance in the research ecosystem," says Peter Henderson, an assistant professor at Princeton University who has worked extensively with open models. If researchers do adopt gpt-oss as new workhorses, OpenAI could see some concrete benefits, Henderson says -- it might adopt innovations discovered by other researchers into its own model ecosystem. More broadly, Lambert says, releasing an open model now could help OpenAI reestablish its status in an increasingly crowded AI environment. "It kind of goes back to years ago, where they were seen as the AI company," he says. Users who want to use open models will now have the option to meet all their needs with OpenAI products, rather than turning to Meta's Llama or Alibaba's Qwen when they need to run something locally. The rise of Chinese open models like Qwen over the past year may have been a particularly salient factor in OpenAI's calculus. An employee from OpenAI emphasized at the media briefing that the company doesn't see these open models as a response to actions taken by any other AI company, but OpenAI is clearly attuned to the geopolitical implications of China's open-model dominance. "Broad access to these capable‬‭ open-weights models created in the US helps expand democratic AI rails," the company wrote in a blog post announcing the models' release.

OpenAI Just Released Its First Open-Weight Models Since GPT-2

The models, gpt-oss-120b and gpt-oss-20b, represent a major shift for the AI company. OpenAI just dropped its first open-weight models in over five years. The two language models, gpt-oss-120b and gpt-oss-20b, can run locally on consumer devices and be fine-tuned for specific purposes. For OpenAI, they represent a shift away from its recent strategy of focusing on proprietary releases, as the company moves towards a wider, and more open, group of AI models that are available for users. "We're excited to make this model, the result of billions of dollars of research, available to the world to get AI into the hands of the most people possible," said OpenAI CEO Sam Altman in an emailed statement. Both gpt-oss-120b and gpt-oss-20b are officially available to download for free on Hugging Face, a popular hosting platform for AI tools. The last open-weight model released by OpenAI was GPT-2, back in 2019. What sets apart an open-weight model is the fact that its "weights" are publicly available, meaning that anyone can peek at the internal parameters to get an idea of how it processes information. Rather than undercutting OpenAI's proprietary models with a free option, cofounder Greg Brockman sees this release as "complementary" to the company's paid services, like the application programming interface currently used by many developers. "Open-weight models have a very different set of strengths," said Brockman in a briefing with reporters. Unlike ChatGPT, you can run a gpt-oss model without a connection to the internet and behind a firewall. Both gpt-oss models use chain-of-thought reasoning approaches, which OpenAI first deployed in its o1 model last fall. Rather than just giving an output, this approach has generative AI tools go through multiple steps to answer a prompt. These new text-only models are not multimodal, but they can browse the web, call cloud-based models to help with tasks, execute code, and navigate software as an AI agent. The smaller of the two models, gpt-oss-20b, is compact enough to run locally on a consumer device with more than 16 GB of memory. The two new models from OpenAI are available under the Apache 2.0 license, a popular choice for open-weight models. With Apache 2.0, models can be used for commercial purposes, redistributed, and included as part of other licensed software. Open-weight model releases from Alibaba's Qwen as well as Mistral also operate under Apache 2.0. Publicly announced in March, the release of these open models was initially delayed for further safety testing. Releasing an open-weight model is potentially more dangerous than a closed-off version since it removes barriers around who can use the tool, and anyone can try to fine-tune a version of gpt-oss for unintended purposes. In addition to the evaluations OpenAI typically runs on its proprietary models, the startup customized the open-weight option to see how it could potentially be misused by a "bad actor" who downloads the tool. "We actually fine-tuned the model internally on some of these risk areas," said Eric Wallace, a safety researcher at OpenAI, "and measured how high we could push them." In OpenAI's tests, the open-weight model did not reach a high level of risk, as measured by its preparedness framework.

OpenAI's New Models Aren't Really Open: What to Know About Open-Weights AI

Despite the company's name, OpenAI hasn't dropped an open version of its AI models since GPT-2 in 2019. That changed on Tuesday, as CEO Sam Altman shared two new open-weights, reasoning AI models, named gpt-oss-120b (120 billion parameters) and gpt-oss-20b (20 billion parameters). If open-weights is a new piece of AI jargon to you, don't worry. In the simplest possible terms, open-weights is a category of AI models that power products like chatbots, image and video generators. But they are philosophically different from the technology underpinning the AI tools you might use now. ChatGPT, Gemini and Copilot are all powered by closed models, which means we have no real insights into how those black box machines work. Open-weights models give us a peek at the mechanical wizard behind the curtain, so to speak. You don't need to be a developer or machine learning expert to understand how these open models work or even to run them yourself. Here's everything you need to know about open-weights and open-source AI models. All AI models have weights, which are characteristics or elements. Models are trained to give certain connections more weight, or value. An open-weights model does exactly what its name implies -- the weights are publicly available, as defined by the Federal Trade Commission. Developers can see these weights and how they're used in the creation of AI models. "Arguably, the most valuable thing in large [language] models is actually the weights. You can do a lot if you have the weights, which is somewhat different from traditional software," Omar Khattab, assistant professor of computer science at MIT and researcher in its computer science and artificial intelligence lab (CSAIL), told CNET. For example, a chatbot is built to be really good at predicting the next logical word in a sentence. It's trained to string together words in its outputs that frequently show up next to each other in its training data, presumably in a logical order. Words that show up next to each other more frequently can be given more weight than words that don't often appear next to each other. These weights are just numbers but open-weights models also come with a map. "In open-weights [models], you get the weights, which are these numbers, and you get how to map those weights into the neural network structure, so the layers of the neural network, in order to actually be able to run it," said Khattab. The architecture of the model shows how a company structures its models, which is "incredibly valuable." Open-weights models are primarily aimed at developers, who can integrate the model into existing projects, like helping to build AI agents. For the "committed hobbyist," as Khattab put it, you can use the specs to run the model locally on your laptop, which could help alleviate privacy concerns that come, for example, with using AI through a company's mobile app. Researchers will also have a clearer look at how the AI works internally. The new open-weights models come in two sizes, 120 billion parameters (128 experts and 128k context window) and 20 billion parameters (32 experts but the same 128k context window). Experts refer to the number of sub-neural networks a model has, and context windows describe how much information a model can process and include in its responses. Bigger numbers for both indicate a model is capable of more sophisticated answers and has more firepower. In terms of performance, OpenAI is reporting that the 120B model "achieved near-parity" with its latest reasoning model, o4-mini on core reasoning benchmarks while running on a single 80 gigabyte GPU. The 20B open-weights model performed similarly to o3-mini and ran on a 16 gigabyte device -- meaning, this smaller open-weights model could be run fairly well on laptops and some smartphones. (Like all AI models run locally, your speed will depend on your device's firepower.) The models will be available under the Apache 2.0 license, a type of open-source-friendly license. You can check out the more in-depth specs in the model card and paper on safety training, get tips from OpenAI's developer guidelines and see the weights for yourself now on HuggingFace and Github. Open-weights models are related to open-source AI, but they aren't exactly the same. Open source as a concept refers to software that has no proprietary owners, whose source code is publicly available and can be used by most anybody under open-source licenses. The Open Source Initiative, a nonprofit that advocates for open-source software, defines open-source AI as "a system made available under terms that grant users the freedom to use, study, modify and share [it]." An open-weights AI model isn't the same as an open-source model. One way to think about the difference between the two is like with baking, Suba Vasudevan, chief operating officer at Mozilla.org and senior vice president at Mozilla corporation, told CNET. "An open-weights model is someone giving you this baked cake and saying, 'Oh, it's made of flour, sugar and eggs.' Those are the weights. But if someone gave you the entire recipe, and all the instructions and the exact details of how much of each ingredient went into it, that's open source," said Vasudevan. For open-weights models, the types of things not disclosed are the data the model was trained on and the code used to train it. Training data is a point of contention between AI companies and the humans creating content; AI companies are ravenous for high-quality, human-generated content to refine and improve their models. Some companies collect this data through licensing agreements, but some publishers and creators have filed lawsuits alleging that AI companies are illegally acquiring their copyrighted content. (Disclosure: Ziff Davis, CNET's parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.) Training data, regardless of its origin, is one of the most valuable things an AI company has. But it likely won't be included in any open-weights model release. "I think because of sheer scale [of data], because of liability, because a lot of it is licensed and you are not supposed to share it, I think it's probably fair to assume that no companies, for-profit at least, are going to be releasing that anytime soon, maybe ever," said Khattab. Truly open-source AI comes with more publicly available information, Vasudevan said. Open-weights models can be harder to retrain or inspect for bias without the additional information. "You're still a fair bit in the dark about how it was built or what data shaped it," Vasudevan said. These differences between open-source, open-weights and closed models likely won't affect your average experience using an AI chatbot. But they're important qualities for developers to consider when selecting and using certain AI models, and they're important more broadly for us to understand the technology that is infiltrating our online lives. There's no guarantee that the inner workings of its open-weights models will reflect what's inside its closed models, Khattab said. But for a company whose product is synonymous with gen AI, any light shed on how it works is sure to have an impact on the people who use it and those who design it behind the scenes. As people start to get into the weeds with the new models, we'll learn more and see what effect it has on the industry going forward. Overall, the underlying philosophy of open source is that technology gets better when more people can access it, like academics who can study it and security experts who can expose weaknesses. "You have to let people build, scale and innovate and be able to poke holes and make it better," said Vasudevan.

OpenAI Releases Open-Weight Models After DeepSeek's Success

OpenAI is releasing a pair of open and freely available artificial intelligence models that can mimic the human process of reasoning, months after China's DeepSeek gained global attention with its own open AI software. The two models, called GPT-oss-120b and GPT-oss-20b, will be available on AI software hosting platform Hugging Face and can produce text -- but not images or videos -- in response to user prompts, OpenAI said on Tuesday. These models can also carry out complex tasks like writing code and looking up information online on a user's behalf, the company said.

OpenAI releases open-weight reasoning models optimized for running on laptops

SAN FRANCISCO, Aug 5 (Reuters) - OpenAI said on Tuesday it has released two open-weight language models that excel in advanced reasoning and are optimized to run on laptops with performance levels similar to its smaller proprietary reasoning models. An open-weight language model's trained parameters or weights are publicly accessible, which can be used by developers to analyze and fine-tune the model for specific tasks without requiring original training data. "One of the things that is unique about open models is that people can run them locally. People can run them behind their own firewall, on their own infrastructure," OpenAI co-founder Greg Brockman said in a press briefing. Open-weight language models are different from open-source models, which provide access to the complete source code, training data and methodologies. The landscape of open-weight and open-source AI models has been highly contested this year. For a time, Meta's (META.O), opens new tab Llama models were considered the best, but that changed earlier this year when China's DeepSeek released a powerful and cost-effective reasoning model, while Meta struggled to deliver Llama 4. The two new OpenAI models are the first open models OpenAI has released since GPT-2, which was released in 2019. OpenAI's larger model, gpt-oss-120b, can run on a single GPU, and the second, gpt-oss-20b, is small enough to run directly on a personal computer, the company said. OpenAI said the models have similar performance to its proprietary reasoning models called o3-mini and o4-mini, and especially excel at coding, competition math and health-related queries. The models were trained on a text-only dataset which in addition to general knowledge, focused on science, math and coding knowledge. OpenAI did not release benchmarks comparing the open-weight models to competitors' models such as the DeepSeek-R1 model. Microsoft-backed OpenAI, currently valued at $300 billion, is currently raising up to $40 billion in a new funding round led by Softbank Group (9984.T), opens new tab. Reporting by Anna Tong in San Francisco; Editing by Edwina Gibbs Our Standards: The Thomson Reuters Trust Principles., opens new tab * Suggested Topics: * Artificial Intelligence Anna Tong Thomson Reuters Anna Tong is a correspondent for Reuters based in San Francisco, where she reports on the technology industry. She joined Reuters in 2023 after working at the San Francisco Standard as a data editor. Tong previously worked at technology startups as a product manager and at Google where she worked in user insights and helped run a call center. Tong graduated from Harvard University.

OpenAI's first new open-weight LLMs in six years are here

For the first time since GPT-2 in 2019, OpenAI is releasing new open-weight large language models. It's a major milestone for a company that has increasingly been accused of forgoing its original stated mission of "ensuring artificial general intelligence benefits all of humanity." Now, following multiple delays for additional safety testing and refinement, gpt-oss-120b and gpt-oss-20b are available to download from Hugging Face. Before going any further, it's worth taking a moment to clarify what exactly OpenAI is doing here. The company is not releasing new open-source models that include the underlying code and data the company used to train them. Instead, it's sharing the weights -- that is, the numerical values the models learned to assign to inputs during their training -- that inform the new systems. According to Benjamin C. Lee, professor of engineering and computer science at the University of Pennsylvania, open-weight and open-source models serve two very different purposes. "An open-weight model provides the values that were learned during the training of a large language model, and those essentially allow you to use the model and build on top of it. You could use the model out of the box, or you could redefine or fine-tune it for a particular application, adjusting the weights as you like," he said. If commercial models are an absolute black box and an open-source system allows for complete customization and modification, open-weight AIs are somewhere in the middle. OpenAI has not released open-source models, likely since a rival could use the training data and code to reverse engineer its tech. "An open-source model is more than just the weights. It would also potentially include the code used to run the training process," Lee said. And practically speaking, the average person wouldn't get much use out of an open-source model unless they had a farm of high-end NVIDIA GPUs running up their electricity bill. (They would be useful for researchers looking to learn more about the data the company used to train its models though, and there are a handful of open-source models out there like Mistral NeMo and Mistral Small 3.) With that out of the way, the primary difference between gpt-oss-120b and gpt-oss-20b is how many parameters each one offers. If you're not familiar with the term, parameters are the settings a large language model can tweak to provide you with an answer. The naming is slightly confusing here, but gpt-oss-120b is a 117 billion parameter model, while its smaller sibling is a 21-billion one. In practice, that means gpt-oss-120b requires more powerful hardware to run, with OpenAI recommending a single 80GB GPU for efficient use. The good news is the company says any modern computer with 16GB of RAM can run gpt-oss-20b. As a result, you could use the smaller model to do something like vibe code on your own computer without a connection to the internet. What's more, OpenAI is making the models available through the Apache 2.0 license, giving people a great deal of flexibility to modify the systems to their needs. Despite this not being a new commercial release, OpenAI says the new models are in many ways comparable to its proprietary systems. The one limitation of the oss models is that they don't offer multi-modal input, meaning they can't process images, video and voice. For those capabilities, you'll still need to turn to the cloud and OpenAI's commercial models, something both new open-weight systems can be configured to do. Beyond that, however, they offer many of the same capabilities, including chain-of-thought reasoning and tool use. That means the models can tackle more complex problems by breaking them into smaller steps, and if they need additional assistance, they know how to use the web and coding languages like Python. Additionally, OpenAI trained the models using techniques the company previously employed in the development of o3 and its other recent frontier systems. In competition-level coding gpt-oss-120b earned a score that is only a shade worse than o3, OpenAI's current state-of-the-art reasoning model, while gpt-oss-20b landed in between o3-mini and o4-mini. Of course, we'll have to wait for more real-world testing to see how the two new models compare to OpenAI's commercial offerings and those of its rivals. The release of gpt-oss-120b and gpt-oss-20b and OpenAI's apparent willingness to double down on open-weight models comes after Mark Zuckerberg signaled Meta would release fewer such systems to the public. Open-sourcing was previously central to Zuckerberg's messaging about his company's AI efforts, with the CEO once remarking about closed-source systems "fuck that." At least among the sect of tech enthusiasts willing to tinker with LLMs, the timing, accidental or not, is somewhat embarrassing for Meta. "One could argue that open-weight models democratize access to the largest, most capable models to people who don't have these massive, hyperscale data centers with lots of GPUs," said Professor Lee. "It allows people to use the outputs or products of a months-long training process on a massive data center without having to invest in that infrastructure on their own. From the perspective of someone who just wants a really capable model to begin with, and then wants to build for some application. I think open-weight models can be really useful." OpenAI is already working with a few different organizations to deploy their own versions of these models, including AI Sweden, the country's national center for applied AI. In a press briefing OpenAI held before today's announcement, the team that worked on gpt-oss-120b and gpt-oss-20b said they view the two models as an experiment; the more people use them, the more likely OpenAI is to release additional open-weight models in the future.

OpenAI releases lower-cost models to rival Meta, Mistral and DeepSeek

OpenAI on Tuesday released two open-weight language models for the first time since it rolled out GPT-2 in 2019. The text-only models are called gpt-oss-120b and gpt-oss-20b, and are designed to serve as lower-cost options that developers, researchers and companies can easily run and customize, OpenAI said. An artificial intelligence model is considered open weight if its parameters, or the elements that improve its outputs and predictions during training, are publicly available. Open-weight models can offer transparency and control, but they are different from open-source models, whose full source code becomes available for people to use and modify. Several other tech companies, including Meta, Microsoft-backed Mistral AI and the Chinese startup DeepSeek, have also released open-weight models in recent years. "It's been exciting to see an ecosystem develop, and we are excited to contribute to that and really push the frontier and then see what happens from there," OpenAI President Greg Brockman told reporters during a briefing. The company collaborated with Nvidia, Advanced Micro Devices, Cerebras, and Groq to ensure the models will work well on a variety of chips. "OpenAI showed the world what could be built on Nvidia AI -- and now they're advancing innovation in open-source software," Nvidia CEO Jensen Huang said in a statement.

OpenAI Finally Lives Up to Its Name, Drops Two New Open Source AI Models

The AI company is looking to get back in the business of transparency. For the first time in five years, OpenAI has released two new free and open-source AI models that are lightweight and designed to be easily integrated into other software programs. In a blog post on Tuesday, the company characterized gpt-oss-120b and gpt-oss-20b as flexible but powerful AI algorithms that can perform a variety of tasks and be used in numerous settings. The company also included a feedback portal and a more extensive blog that further explains the models and how they work. OpenAI's CEO, Sam Altman, said via X on Tuesday that he hoped that the AI would assist with "new kinds of research and the creation of new kinds of products." Altman also seemed to champion the open source method: "We believe in individual empowerment," he said. "Although we believe most people will want to use a convenient service like ChatGPT, people should be able to directly control and modify their own AI when they need to, and the privacy benefits are obvious." Unlike most of ChatGPT's products, open-source models disclose the training parameters that were used to build their system. This level of transparency affords onlookers the benefit of knowing how the system functions and why it might behave in the ways that it does. The last time OpenAI released an open-source model was during Trump's first presidency (truly, that feels like decades ago), with the release of GPT-2. That was when OpenAI was still a little-known startup, and it would be several years before the launch of ChatGPT in 2022. Back in those days, the company was still routinely being referred to by monikers like "Elon Musk's AI project," despite the fact that Musk had parted ways with the company. Perhaps most importantly, you won't have to pay a dime to use these models. And as long as your computer's specs are up to snuff, you can run them locally instead of relying on OpenAI's servers. Any preppers out there looking for an AI doomsday machine might want to look into it. From the looks of things, there's a lot of promising stuff in the company's new releases. gpt-oss is built to integrate into agentic workflows, OpenAI says, which means that new types of automated workâ€"conducted by so-called "agents"â€"can be powered by the new algorithms. The new models also fall under the Apache 2.0 license, which allows users to create new software with the algorithms without worrying about getting sued. "Build freely without worrying about copyleft restrictions or patent riskâ€"whether you're experimenting, customizing, or deploying commercially," OpenAI writes. The open-source ecosystem largely subsists on a plethora of such licensing agreements, allowing companies to build off free models. OpenAI also paid some lip service to AI safety in its announcement. The company claims that, in addition to "running the models through comprehensive safety training and evaluations," it also "introduced an additional layer of evaluation by testing an adversarially fine-tuned version of gpt-oss-120b" using its Preparedness Framework, which is designed to assess and track risky behavior in large language models. In recent years, much criticism has been aimed at OpenAI over its decision to continue the "walled garden" approach to software development. The company's LLM releases have stayed proprietary and, thus, shut off from public inspection. Now, OpenAI is obviously trying to prove the haters wrong and double down on its commitment to being a truly "open" organization. It's anybody's guess as to whether the company will be able to maintain such a commitment to the FOSS ethos, given that it also happens to be an organization worth hundreds of billions of dollars on paper. The organized money behind OpenAI may continue to see a benefit in owning systems that are exclusive, closed, and, most importantly, only controlled by a select group of insiders. It's noteworthy that GPT-5, the company's most powerful and highly anticipated new model, will almost certainly be released in the same fashion as other recent GPT releases: closed.

OpenAI releases two open-weight AI models, including one that runs well on Apple Silicon Macs - 9to5Mac

Living up to its name, OpenAI has released not one but two new open-weight AI models after promising to deliver a new open-weight model earlier this year. The two models, gpt-oss-20b and gpt-oss-120b, are available to download for free now. What makes open-weight models special? Specifically, these are AI models that can be downloaded and run on computers with adequate resources for powering local AI models. No internet connection is required because access to the model provider's server isn't involved. This also allows developers to build custom tools using the open AI models. OpenAI describes the '20b' variant as a medium-sized open model while it calls the '120b' variant a large-sized open model for running on "most desktops and laptops." OpenAI says the smaller model works best with at least 16GB VRAM or unified memory and is "perfect for higher-end consumer GPUs and Apple Silicon Macs." The larger, full-sized model wants at least 60GB VRAM or unified memory. While these models aren't the most powerful AI tools from OpenAI, they should satisfy the demand for open-weight models from the team behind ChatGPT. The rise of DeepSeek out of China last year put pressure on OpenAI to deliver a modern open-weight model. Prior to today, OpenAI hadn't released an open-weight model since its early days in 2019.

OpenAI releases gpt-oss, new open-weight models that can run on laptops: How to try them

For the first time since GPT-2 dropped over five years ago, OpenAI is releasing not one, but two open-weight AI reasoning models -- and they're available to download for free right now on Hugging Face. Billed as "state-of-the-art," the new OpenAI open-weight models were announced Tuesday in a company blog post. OpenAI says they "outperform similarly sized open models on reasoning, excel at tool use, and are optimized for efficient deployment on consumer hardware." This Tweet is currently unavailable. It might be loading or has been removed. There are two versions: To try out the new OpenAI models for yourself, head to the OpenAI gpt-oss page. This release is a nod to OpenAI's early roots, when the company was more publicly committed to open-sourcing its models (hence the company's name). While these aren't "fully open source" in the strictest sense -- the training data isn't included -- they are open weight, meaning the code and model parameters are available for anyone to use, tweak, and build upon. And no, models like Meta's LLaMA aren't truly open source either -- at least not by the standards of the open-source community, which requires access to training data as a baseline. Since the release of GPT-2, OpenAI has steadily shifted toward a more closed and proprietary approach to its LLM development -- until now. The recent release of open-weight models marks a notable change in direction, and it's not happening in a vacuum. With China's DeepSeek AI and other labs in the country gaining traction and achieving impressive scores on benchmark tests, the pressure has been mounting for US tech companies to stay competitive in the global AI race. In fact, just last month, the Trump administration urged American AI developers to open source more of their technology in an effort to promote innovation aligned with "American values" and maintain a strategic edge. Regardless of the motivations behind it, this move represents a significant step forward, not just for OpenAI but for the broader open AI ecosystem.

OpenAI just announced something big, but it's not GPT-5

These models let users see each step of the AI's thinking process, making them more transparent in how they function OpenAI has just dropped two new AI models, gpt‑oss‑120b and gpt‑oss‑20b. Not only are they new, but they're the first open‑weight models from ChatGPT's creator since GPT‑2. The smaller of the two - gpt-oss-20b - is especially notable for being light enough to run on a decently specced consumer PC. If you've got about 16GB of RAM and some patience, you can load it up, ask it questions, and actually see how it arrives at answers. The larger 120b model still requires serious hardware or cloud support. Both models are part of OpenAI's new push to encourage developers to play around with the models and even commercialize them for the average user. For the first time in years, developers and curious individuals alike can download and run OpenAI models on their own machines, inspect how they think, and build on them freely. They're available via Hugging Face and AWS under the Apache 2.0 license. Being open weight means the models provide a level of transparency and independence that most people haven't had since ChatGPT first went viral. The real-time reasoning is visible throughout, and you can see how the 'logic' of the model leads to its final options for how to respond and how it makes that decision. That's a big shift for OpenAI. The company has spent several years restricting access to its most powerful tools, offering only API endpoints and paid tiers. Now it's returning a little bit to the GPT-2 era, but with far more capable models. Even so, the lighter model isn't something everyone will rush to as a replacement for the ChatGPT app. The flexibility provided by the new models could be a boon for OpenAI as the open-weight approach becomes more popular. DeepSeek, Meta, and Mistral have all released open models in some fashion recently. But most have turned out to be semi-open, meaning they are trained on undisclosed data or have constricted terms and usage limits. The gpt-oss models are straightforward in offering the weights and license, though the training data remains proprietary. And OpenAI's gpt-oss models bring compatibility with OpenAI's widely used interface, as well as a bigger window into how the model makes decisions that stand out. So, where DeepSeek models tend to emphasize the raw power and relatively low-cost performance of their models, OpenAI is more interested in explaining how the models work. That will pique the interest of many developers trying to learn or experiment. You can literally pause and look at what's going on inside the model. It's also a signal to the developer community: OpenAI is inviting people to build with its tech in a way that hasn't been possible for years. These models don't just output chat completions; they offer a foundation. With the Apache license, you can fine-tune, distill, embed, or wrap them into entirely new products. And you don't need to send data to a third party to do it. Those hoping for a more decentralized AI ecosystem are likely to be pleased. For those of us who are less technical, the main point is that you might see AI apps that run very well without needing a subscription price from you, or personalized tools that don't send your data to the cloud. With this release, OpenAI can claim to be willing to share more with developers, if not quite everything. For a company that's spent much of the past two years building closed systems and premium subscriptions, releasing a pair of models this capable under open licenses feels like a major philosophical change, or at least the desire to make such a change seem real. And while these two models don't represent the push into the era of GPT-5, OpenAI chief Sam Altman has teased that even more announcements are on the horizon, so it's likely that the next-generation model is still being readied.

OpenAI returns to open source roots with new models gpt-oss-120b and gpt-oss-20b

Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders. Subscribe Now OpenAI is getting back to its roots as an open source AI company with today's announcement and release of two new, open source, frontier large language models (LLMs): gpt-oss-120b and gpt-oss-20b. The former is a 120-billion parameter model as the name would suggest, capable of running on a single Nvidia H100 graphics processing unit (GPU) and the latter is only 20 billion, small enough to run locally on a consumer laptop or desktop PC. Both are text-only language models, which means unlike the multimodal AI that we've had for nearly two years that allows users to upload files and images and have the AI analyze them, users will be confined to only inputting text messages to the models and receiving text back out. However, they can still of course write code and provide math problems and numerics, and in terms of their performance on tasks, they rank above some of OpenAI's paid models and much of the competition globally. They can also be connected to external tools including web search to perform research on behalf of the user. More on this below. Most importantly: they're free, they're available for enterprises and indie developers to download the code and use right now, modifying according to their needs, and can be run locally without a web connection, ensuring maximum privacy, unlike the other top OpenAI models and those from leading U.S.-based rivals Google and Anthropic. The models can be downloaded today with full weights (the settings guiding its behavior) on the AI code sharing community Hugging Face and GitHub. High benchmark scores According to OpenAI, gpt-oss-120b matches or exceeds its proprietary o4-mini model on reasoning and tool-use benchmarks, including competition mathematics (AIME 2024 & 2025), general problem solving (MMLU and HLE), agentic evaluations (TauBench), and health-specific evaluations (HealthBench). The smaller gpt-oss-20b model is comparable to o3-mini and even surpasses it in some benchmarks. The models are multilingual and perform well across a variety of non-English languages, though OpenAI declined to specify which and how many. While these capabilities are available out of the box, OpenAI notes that localized fine-tuning -- such as an ongoing collaboration with the Swedish government to produce a version fine-tuned on the country's language -- can still meaningfully enhance performance for specific regional or linguistic contexts. A hugely advantageous license for enterprises and privacy-minded users But the biggest feature is the licensing terms for both: Apache 2.0, the same as the wave of Chinese open source models that have been released over the last several weeks, and a more enterprise-friendly license than Meta's trickier and more nuanced open-ish Llama license, which requires that users who operate a service with more than 700 million monthly active users obtain a paid license to keep using the company's family of LLMs. By contrast, OpenAI's new gpt-oss series of models offer no such restrictions. In keeping with Chinese competitors and counterparts, any consumer, developer, independent entrepreneur or enterprise large and small is empowered by the Apache 2.0 license to be able to download the new gpt-oss models at will, fine-tune and alter them to fit their specific needs, and use them to generate revenue or operate paid services, all without paying OpenAI a dime (or anything!). This also means enterprises can use a powerful, near topline OpenAI model on their own hardware totally privately and securely, without sending any data up to the cloud, on web servers, or anywhere else. For highly regulated industries like finance, healthcare, and legal services, not to mention organizations in military, intelligence, and government, this may be a requirement. Before today, anyone using ChatGPT or its application programming interface (API) -- the service that acts like a switching board and allows third-party software developers to connect their own apps and services to these OpenAI's proprietary/paid models like GPT-4o and o3 -- was sending data up to OpenAI servers that could technically be subpoenaed by government agencies and accessed without a user's knowledge. That's still the case for anyone using ChatGPT or the API going forward, as OpenAI co-founder and Sam Altman recently warned. And while running the new gpt-oss models locally on a user's own hardware disconnected from the web would allow for maximum privacy, as soon as the user decides to connect it to external web search or other web enabled tools, some of the same privacy risks and issues would then arise -- through any third-party web services the user or developer was relying on when hooking the models up to said tools. The last OpenAI open source language model was released more than six years ago "This is the first time we're releasing an open-weight language model in a long time... We view this as complementary to our other products," said OpenAI co-founder and president Greg Brockman on an embargoed press video call with VentureBeat and other journalists last night. The last time OpenAI released a fully open source language model was GPT-2 in 2019, more than six years ago, and three years before the release of ChatGPT. This fact has sparked the ire of -- and resulted in several lawsuits from -- former OpenAI co-founder and backer turned rival Elon Musk, who, along with many other critics, have spent the last several years accusing OpenAI of betraying its mission and founding principles and namesake by eschewing open source AI releases in favor of paid proprietary models available only to customers of OpenAI's API or paying ChatGPT subscribers (though there is a free tier for the latter). OpenAI co-founder CEO Sam Altman did express regret about being on the "wrong side of history" but not releasing more open source AI sooner in a Reddit AMA (ask me anything) QA with users in February of this year, and Altman committed to releasing a new open source model back in March, but ultimately the company delayed its release from a planned July date until now. Now OpenAI is tacking back toward open source, and the question is, why? Why would OpenAI release a set of free open source models that it makes no money from? To paraphrase Jesse Plemons' character's memorable line from the film Game Night: "How can that be profitable for OpenAI?" After all, business to OpenAI's paid offerings appears to be booming. Revenue has skyrocketed alongside the rapid expansion of its ChatGPT user base, now at 700 million weekly active users. As of August 2025, OpenAI reported $13 billion in annual recurring revenue, up from $10 billion in June. That growth is driven by a sharp rise in paying business customers -- now 5 million, up from 3 million just two months earlier -- and surging daily engagement, with over 3 billion user messages sent every day. The financial momentum follows an $8.3 billion funding round that valued OpenAI at $300 billion and provides the foundation for the company's aggressive infrastructure expansion and global ambitions. Compare that to closed/proprietary rival AI startup Anthropic's reported $5 billion in total annual recurring revenue, but interestingly, Anthropic is said to be getting more money from its API, $3.1 billion in revenue compared to OpenAI's $2.9 billion, according to The Information. So, given how well the paid AI business is already doing, the business strategy behind these open source offerings is less clear -- especially since the new OpenAI gpt-oss models will almost certainly cut into some (perhaps a lot of) usage of OpenAI's paid models. Why go back to offering open source LLMs now when so much money is flowing into paid and none will, by virtue of its very intent, go directly toward open source models? Put simply: because open source competitors, beginning with the release of the impressively efficient DeepSeek R1 by the Chinese AI division of the same name in January 2025, are offering near parity on performance benchmarks to paid proprietary models, for free, with fewer (basically zero) implementation restrictions for enterprises and end users. And increasingly, enterprises are adopting these open source models in production. As OpenAI executives and team members revealed to VentureBeat and many other journalists on an embargoed video call last night about the new models that when it comes to OpenAI's API, the majority of customers are using a mix of paid OpenAI models and open source models from other providers. (I asked, but OpenAI declined to specify what percentage or total number of API customers are using open source models and which ones). At least, until now. OpenAI clearly hopes these new gpt-oss offerings will get more of these users to switch away from competing open source offerings and back into OpenAI's ecosystem, even if OpenAI doesn't see any direct revenue or data from that usage. On a grander scale, it seems OpenAI wants to be a full-service, full-stack, one-stop shop AI offering for all of an enterprise, indie developer's, or regular consumer's machine intelligence needs -- from a clean chatbot interface to an API to build services and apps atop of to agent frameworks for building AI agents through said API to an image generation model (gpt-4o native image generation), video model (Sora), audio transcription model (gpt-4o-transcribe), and now, open source offerings as well. Can a music generation and "world model" be far behind? OpenAI seeks to span the AI market, propriety and open source alike, even if the latter is worth nothing in terms of actual, direct dollars and cents. Training and architecture Feedback from developers directly influenced gpt-oss's design. OpenAI says the top request was for a permissive license, which led to the adoption of Apache 2.0 for both models. Both models use a Mixture-of-Experts (MoE) architecture with a Transformer backbone. The larger gpt-oss-120b activates 5.1 billion parameters per token (out of 117 billion total), and gpt-oss-20b activates 3.6 billion (out of 21 billion total). Both support 128,000 token context length (about 300-400 pages of a novel's worth of text a user can upload at once), and employ locally banded sparse attention and use Rotary Positional Embeddings for encoding. The tokenizer -- the program that converts words and chunks of words into the numerical tokens that the LLMs can understand, dubbed "o200k_harmony" -- is also being open-sourced. Developers can select among low, medium, or high reasoning effort settings based on latency and performance needs. While these models can reason across complex agentic tasks, OpenAI emphasizes they were not trained with direct supervision of CoT outputs, to preserve the observability of reasoning behavior -- an approach OpenAI considers important for safety monitoring. Another common request from OpenAI's developer community was for strong support for function calling, particularly for agentic workloads, which OpenAI believes gpt-oss now delivers. The models are engineered for chain-of-thought reasoning, tool use, and few-shot function calling, and are compatible with OpenAI's Responses API introduced back in March, which allows developers to augment their apps by connecting an OpenAI LLM of their choice to three powerful built-in tools -- web search, file search, and computer use -- within a single API call. But for the new gpt-oss models, tool use capabilities -- including web search and code execution -- are not tied to OpenAI infrastructure. OpenAI provides the schemas and examples used during training, such as a basic browser implementation using the Exa API and a Python interpreter that operates in a Docker container. It is up to individual inference providers or developers to define how tools are implemented. Providers like vLLM, for instance, allow users to configure their own MCP (Model-Controller-Proxy) server to specify the browser backend. While these models can reason across complex agentic tasks, OpenAI emphasizes they were not trained with direct supervision of CoT outputs, to preserve the observability of reasoning behavior -- an approach OpenAI considers important for safety monitoring. Safety evaluations and measures OpenAI conducted safety training using its Preparedness Framework, a document that outlines the procedural commitments, risk‑assessment criteria, capability categories, thresholds, evaluations, and governance mechanisms OpenAI uses to monitor, evaluate, and mitigate frontier AI risks. These included filtering chemical, biological, radiological, and nuclear threat (CBRN) related data out during pretraining, and applying advanced post-training safety methods such as deliberative alignment and an instruction hierarchy to enforce refusal behavior on harmful prompts. To test worst-case misuse potential, OpenAI adversarially fine-tuned gpt-oss-120b on sensitive biology and cybersecurity data using its internal RL training stack. These malicious fine-tuning (MFT) scenarios -- one of the most sophisticated evaluations of this kind to date -- included enabling browsing and disabling refusal behavior, simulating real-world attack potential. The resulting models were benchmarked against both open and proprietary LLMs, including DeepSeek R1-0528, Qwen 3 Thinking, Kimi K2, and OpenAI's o3. Despite enhanced access to tools and targeted training, OpenAI found that even the fine-tuned gpt-oss models remained below the "High" capability threshold for frontier risk domains such as biorisk and cybersecurity. These conclusions were reviewed by three independent expert groups, whose recommendations were incorporated into the final methodology. In parallel, OpenAI partnered with SecureBio to run external evaluations on biology-focused benchmarks like Human Pathogen Capabilities Test (HPCT), Molecular Biology Capabilities Test (MBCT), and others. Results showed that gpt-oss's fine-tuned models performed close to OpenAI's o3 model, which is not classified as frontier-high under OpenAI's safety definitions. According to OpenAI, these findings contributed directly to the decision to release gpt-oss openly. The release is also intended to support safety research, especially around monitoring and controlling open-weight models in complex domains. Availability and ecosystem support The gpt-oss models are now available on Hugging Face, with pre-built support through major deployment platforms including Azure, AWS, Databricks, Cloudflare, Vercel, Together AI, OpenRouter, and others. Hardware partners include NVIDIA, AMD, and Cerebras, and Microsoft is making GPU-optimized builds available on Windows via ONNX Runtime. OpenAI has also announced a $500,000 Red Teaming Challenge hosted on Kaggle, inviting researchers and developers to probe the limits of gpt-oss and identify novel misuse pathways. A public report and an open-source evaluation dataset will follow, aiming to accelerate open model safety research across the AI community. Early adopters such as AI Sweden, Orange, and Snowflake have collaborated with OpenAI to explore deployments ranging from localized fine-tuning to secure on-premise use cases. OpenAI characterizes the launch as an invitation for developers, enterprises, and governments to run state-of-the-art language models on their own terms. While OpenAI has not committed to a fixed cadence for future open-weight releases, it signals that gpt-oss represents a strategic expansion of its approach -- balancing openness with aligned safety methodologies to shape how large models are shared and governed in the years ahead. The big question: with so much competition in open source AI, will OpenAI's own efforts pay off? OpenAI re-enters the open source model market in the most competitive moment yet. At the top of public AI benchmarking leaderboards, U.S. frontier models remain proprietary -- OpenAI (GPT-4o/o3), Google (Gemini), and Anthropic (Claude). But they now compete directly with a surge of open-weights contenders. From China: DeepSeek-R1 (open source, MIT) and DeepSeek-V3 (open-weights under a DeepSeek Model License that permits commercial use); Alibaba's Qwen 3 (open-weights, Apache-2.0); MoonshotAI's Kimi K2 (open-weights; public repo and model cards); and Z.ai's GLM-4.5 (also Apache 2.0 licensed). Europe's Mistral (Mixtral/Mistral, open-weights, Apache-2.0) anchors the EU push; the UAE's Falcon 2/3 publish open-weights under TII's Apache-based license. In the U.S. open-weights camp, Meta's Llama 3.1 ships under a community (source-available) license, Google's Gemma under Gemma terms (open weights with use restrictions), and Microsoft's Phi-3.5 under MIT. Developer pull mirrors that split. On Hugging Face, Qwen2.5-7B-Instruct (open-weights, Apache-2.0) sits near the top by "downloads last month," while DeepSeek-R1 (MIT) and DeepSeek-V3 (model-licensed open weights) also post heavy traction. Open-weights stalwarts Mistral-7B / Mixtral (Apache-2.0), Llama-3.1-8B/70B (Meta community license), Gemma-2 (Gemma terms), Phi-3.5 (MIT), GLM-4.5 (open-weights), and Falcon-2-11B (TII Falcon License 2.0) round out the most-pulled families -- underscoring that the open ecosystem spans the U.S., Europe, the Middle East, and China. Hugging Face signals adoption, not market share, but they show where builders are experimenting and deploying today. Consumer usage remains concentrated in proprietary apps even as weights open up. ChatGPT still drives the largest engagement globally (about 2.5 billion prompts/day, proprietary service), while in China the leading assistants -- ByteDance's Doubao, DeepSeek's app, Moonshot's Kimi, and Baidu's ERNIE Bot -- are delivered as proprietary products, even as several base models (GLM-4.5, ERNIE 4.5 variants) now ship as open-weights. But now that a range of powerful open source models are available to businesses and consumers -- all nearing one another in terms of performance -- and can be downloaded on consumer hardware, the big question facing OpenAI is: who will pay for intelligence at all? Will the convenience of the web-based chatbot interface, multimodal capabilities, and more powerful performance be enough to keep the dollars flowing? Or has machine intelligence already become, in the words of Atlman himself, "too cheap to meter"? And if so, how to build a successful business atop it, especially with OpenAI and other AI firms' sky-high valuations and expenditures. One clue: OpenAI is already said to be offering in-house engineers to help its enterprise customers customize and deploy fine-tuned models, similar to Palantir's "forward deployed" software engineers (SWEs), essentially charging for experts to come in, set up the models correctly, and train employees how to use them for best results. Perhaps the world will migrate toward a majority of AI usage going to open source models, or a sizeable minority, with OpenAI and other AI model providers offering experts to help install said models into enterprises. Is that enough of a service to build a multi-billion dollar business upon? Or will enough people continue paying $20, $200 or more each month to have access to even more powerful proprietary models? I don't envy the folks at OpenAI figuring out all the business calculations -- despite what I assume to be hefty compensation as a result, at least for now. But for end users and enterprises, the release of the gpt-oss series is undoubtedly compelling.

OpenAI releases powerful new open language models

Why it matters: OpenAI is aiming the new models at customers who want the cost savings and privacy benefits that come from running AI models directly on their own devices rather than relying on cloud-based services like ChatGPT or its rivals. * It's also pitching the open models for countries that want to avoid getting their AI tools from the cloud servers of Google, Microsoft or other tech giants. The big picture: The arrival of China's DeepSeek earlier this year jolted the open-model world and suggested that China might be taking the lead in that category, while Meta's commitment to its open source Llama project has come into question as the company pivots to the "superintelligence" race. What they're saying: "We're excited to make this model, the result of billions of dollars of research, available to the world to get AI into the hands of the most people possible," CEO Sam Altman said in a statement. * "Going back to when we started in 2015, OpenAI's mission is to ensure AGI that benefits all of humanity," Altman said. "To that end, we are excited for the world to be building on an open AI stack created in the United States, based on democratic values, available for free to all and for wide benefit." Driving the news: OpenAI is releasing two new open models, both capable of chain-of-thought reasoning and accessing the web. They can also, if desired, work in conjunction with larger cloud-based AI models. * The first, a 117 billion parameter model called gpt-oss-120b, can run on a single GPU with 80 gigabytes of RAM. * The second, with 21 billion parameters called gpt-oss-20b, is designed to run on laptops or other devices with 16 gigabytes of RAM. * Both models are available via Hugging Face and other cloud providers. Microsoft is also making available a version of the smaller model that has been optimized to run on Windows devices. * The company provided various benchmarks showing the open models performing at or near the performance of the company's o3 and o4-mini models. Yes, but: The new open-models are text-only, as compared to most of OpenAI's recent models, which are so-called multimodal models capable of processing and outputting text, images, audio and video. Between the lines: Technically, the models are "open weights" versus "open source," meaning anyone can download and fine-tune the models but there's no public access to other key information, like training data details.

OpenAI releases free, downloadable models in competition catch-up

OpenAI on Tuesday released two new artificial intelligence (AI) models that can be downloaded for free and altered by users, to challenge similar offerings by US and Chinese competition. The release of gpt-oss-120b and gpt-oss-20b "open-weight language models" comes as the ChatGPT-maker is under pressure to share inner workings of its software in the spirit of its origin as a nonprofit. "Going back to when we started in 2015, OpenAI's mission is to ensure AGI (Artificial General Intelligence) that benefits all of humanity," said OpenAI chief executive Sam Altman. An open-weight model, in the context of generative AI, is one in which the trained parameters are made public, enabling users to fine-tune it. Meta touts its open-source approach to AI, and Chinese AI startup DeepSeek rattled the industry with its low-cost, high-performance model boasting an open weight approach that allows users to customize the technology. "This is the first time that we're releasing an open-weight model in language in a long time, and it's really incredible," OpenAI co-founder and president Greg Brockman said during a briefing with journalists. The new, text-only models deliver strong performance at low cost, according to OpenAI, which said they are suited for AI jobs like searching the internet or executing computer code, and are designed to be easy to run on local computer systems. "We are quite hopeful that this release will enable new kinds of research and the creation of new kinds of products," Altman said. OpenAI said it is working with partners including French telecommunications giant Orange and cloud-based data platform Snowflake on real-world uses of the models. The open-weight models have been tuned to thwart being used for malicious purposes, according to OpenAI. Altman early this year said his company had been "on the wrong side of history" when it came to being open about how its technology works. He later announced that OpenAI will continue to be run as a nonprofit, abandoning a contested plan to convert into a for-profit organization. The structural issue had become a point of contention, with major investors pushing for better returns. That plan faced strong criticism from AI safety activists and co-founder Elon Musk, who sued the company he left in 2018, claiming the proposal violated its founding philosophy. In the revised plan, OpenAI's money-making arm will be open to generate profits but will remain under the nonprofit board's supervision.

OpenAI's new 'open' models are here -- just don't expect them to reveal the secret recipe

Despite what its name suggests, OpenAI hadn't released an "open" model -- one that includes access to the weights, or the numerical parameters often described as the model's brains -- since GPT-2 in 2020. That changed on Tuesday: The company launched a long-awaited open-weight model, in two sizes, that the company says pushes the frontier of reasoning in open-source AI. "We're excited to make this model, the result of billions of dollars of research, available to the world to get AI into the hands of the most people possible," said OpenAI CEO Sam Altman about the release. "As part of this, we are quite hopeful that this release will enable new kinds of research and the creation of new kinds of products." He emphasized that he is "excited for the world to be building on an open AI stack created in the United States, based on democratic values, available for free to all and for wide benefit." OpenAI CEO Sam Altman had teased the upcoming models back in March, two months after admitting, in the wake of the success of China's open models from DeepSeek, that the company had been "on the wrong side of history" when it came to opening up its models to developers and builders. But while the weights are now public, experts note that OpenAI's new models are hardly "open." By no means is it giving away its crown jewels: The proprietary architecture, routing mechanisms, training data and methods that power its most advanced models -- including the long-awaited GPT-5, widely expected to be released sometime this month -- remain tightly under wraps. The two new model names - gpt-oss-120b and gpt-oss-20b - may be indecipherable to non-engineers, but that's because OpenAI is setting its sights on AI builders and developers seeking to rapidly build on real-world use cases on their own systems. The company noted that the larger of the two models can run on a single Nvidia 80GB chip while the smaller one fits on consumer hardware like a Mac laptop. Greg Brockman, co-founder and president of OpenAI, acknowledged on a press pre-briefing call that "it's been a long time" since the company had released an open model. He added that it is "something that we view as complementary to the other products that we release" and along with OpenAI's proprietary models, "combine to really accelerate our mission of ensuring that API benefits all of humanity." OpenAI said the new models perform well on reasoning benchmarks, which have emerged as the key measurements for AI performance, with models from OpenAI, Anthropic, Google and DeepSeek fiercely competing over their abilities to tackle multi-step logic, code generation, and complex problem-solving. Ever since the open source DeepSeek R1 shook the industry in January with its reasoning capabilities at a much lower cost, many other Chinese models have followed suit- including Alibaba's Qwen and Moonshot AI's Kimi models. While OpenAI said at a press pre-briefing that the new open-weight models are a proactive effort to provide what users want, it is also clearly a strategic response to ramping up open source competition. Notably, OpenAI declined to benchmark its new open-weight models against Chinese open-source systems like DeepSeek or Qwen -- despite the fact that those models have recently outperformed U.S. rivals on key reasoning benchmarks. In the press briefing, the company said it confident in its benchmarks against its own models and that it would leave it to others in the AI community to test further and "make up their own minds." OpenAI's new open-weight models are built using a Mixture-of-Experts (MoE) architecture, in which the system activates only the "experts," or sub-networks, it needs for a specific input, rather than using the entire model for every query. Dylan Patel, founder of research firm SemiAnalysis, pointed out in a post on X before the release that OpenAI trained the models only using publicly known components of the architecture -- meaning the building blocks it used are already familiar to the open-source community. He emphasized that this was a deliberate choice -- that by avoiding any proprietary training techniques or architecture innovations, OpenAI could release a genuinely useful model without actually leaking any intellectual property that powers its proprietary frontier models like GPT 4o. For example, in a model card accompanying the release, OpenAI confirmed that the models use a Mixture of Experts (MoE) architecture with 12 active experts out of 64, but it does not describe the routing mechanism, which is a crucial and proprietary part of the architecture. "You want to minimize risk to your business, but you [also] want to be maximally useful to the public," Aleksa Gordic, a former Google DeepMind researcher, told Fortune, adding that companies like Meta and Mistral, which have also focused on open-weight models, also have not included proprietary information. "They minimize the IP leak and remove any risk to their core business, while at the same time sharing a useful artifact that will enable the startup ecosystem and developers," he said. "It's by definition the best they can do given those two opposing objectives."

OpenAI Drops Two Open Source AI Models That Run Locally and Match Premium Offerings - Decrypt

Performance rivals GPT-4o-mini and o3 -- beats rivals in math, code, and medical benchmarks OpenAI released two open-weight language models Tuesday that deliver performance matching its commercial offerings while running on consumer hardware -- the gpt-oss-120b needs a single 80GB GPU and the gpt-oss-20b operates on devices with just 16GB of memory. The models, available under Apache 2.0 licensing, achieve near-parity with OpenAI's o4-mini on reasoning benchmarks. The 120-billion parameter version activates only 5.1 billion parameters per token through its mixture-of-experts architecture, while the 20-billion parameter model activates 3.6 billion. Both handle context lengths up to 128,000 tokens -- the same as GPT-4o. The fact that they are released under that specific license is a pretty big deal. It means anyone can use, modify and profit from those models without restrictions. This includes anyone from you to OpenAI's competitors like Chinese startup DeepSeek. The release comes as speculation mounts about GPT-5's imminent arrival and competition intensifies in the open-source AI space. The OSS models are OpenAI's latest open-weight language models since GPT-2 in 2019. There is not really a release date for GPT-5, but Sam Altman hinted it could happen sooner than later. "We have a lot of new stuff for you over the next few days," he tweeted early today, promising "a big upgrade later this week." The open-source models that dropped today are very powerful. "These models outperform similarly sized open models on reasoning tasks, demonstrate strong tool use capabilities, and are optimized for efficient deployment on consumer hardware," OpenAI stated in its announcement. The company trained them using reinforcement learning and techniques from its o3 and other frontier systems. On Codeforces competition coding, gpt-oss-120b scored an Elo rating of 2622 with tools and 2463 without -- surpassing o4-mini's 2719 rating and approaching o3's 2706. The model hit 96.6% accuracy on AIME 2024 mathematics competitions compared to o4-mini's 87.3% and achieved 57.6% on the HealthBench evaluation, beating o3's 50.1% score. The smaller gpt-oss-20b matched or exceeded o3-mini across these benchmarks despite its size. It scored 2516 Elo on Codeforces with tools, reached 95.2% on AIME 2024, and hit 42.5% on HealthBench -- all while fitting in memory constraints that would make it viable for edge deployment. Both models support three reasoning effort levels -- low, medium, and high -- that trade latency for performance. Developers can adjust these settings with a single sentence in the system message. The models were post-trained using processes similar to o4-mini, including supervised fine-tuning and what OpenAI described as a "high-compute RL stage." But don't think just because anyone can modify those models at will, you'll have an easy time. OpenAI filtered out certain harmful data related to chemical, biological, radiological, and nuclear threats during pre-training. The post-training phase used deliberative alignment and instruction hierarchy to teach refusal of unsafe prompts and defense against prompt injections. In other words, OpenAI claims to have designed its models to make them so safe, they cannot generate harmful responses even after modifications. Eric Wallace, an OpenAI alignment expert, revealed the company conducted unprecedented safety testing before release. "We fine-tuned the models to intentionally maximize their bio and cyber capabilities," Wallace posted on X. The team curated domain-specific data for biology and trained the models in coding environments to solve capture-the-flag challenges. The adversarially fine-tuned versions underwent evaluation by three independent expert groups. "On our frontier risk evaluations, our malicious-finetuned gpt-oss underperforms OpenAI o3, a model below Preparedness High capability," Wallace stated. The testing indicated that even with robust fine-tuning using OpenAI's training stack, the models couldn't reach dangerous capability levels according to the company's Preparedness Framework. That said, the models maintain unsupervised chain-of-thought reasoning, which OpenAI said is of paramount importance for keeping a wary eye on the AI. "We did not put any direct supervision on the CoT for either gpt-oss model," the company stated. "We believe this is critical to monitor model misbehavior, deception and misuse." OpenAI hides the full chain of thought on its best models to prevent competition from replicating their results -- and to avoid another DeepSeek event, which now can happen even easier.

OpenAI releases free, downloadable models in competition catch-up

San Francisco (United States) (AFP) - OpenAI on Tuesday released two new artificial intelligence (AI) models that can be downloaded for free and altered by users, to challenge similar offerings by US and Chinese competition. The release of gpt-oss-120b and gpt-oss-20b "open-weight language models" comes as the ChatGPT-maker is under pressure to share inner workings of its software in the spirit of its origin as a nonprofit. "Going back to when we started in 2015, OpenAI's mission is to ensure AGI (Artificial General Intelligence) that benefits all of humanity," said OpenAI chief executive Sam Altman. An open-weight model, in the context of generative AI, is one in which the trained parameters are made public, enabling users to fine-tune it. Meta touts its open-source approach to AI, and Chinese AI startup DeepSeek rattled the industry with its low-cost, high-performance model boasting an open weight approach that allows users to customize the technology. "This is the first time that we're releasing an open-weight model in language in a long time, and it's really incredible," OpenAI co-founder and president Greg Brockman said during a briefing with journalists. The new, text-only models deliver strong performance at low cost, according to OpenAI, which said they are suited for AI jobs like searching the internet or executing computer code, and are designed to be easy to run on local computer systems. "We are quite hopeful that this release will enable new kinds of research and the creation of new kinds of products," Altman said. OpenAI said it is working with partners including French telecommunications giant Orange and cloud-based data platform Snowflake on real-world uses of the models. The open-weight models have been tuned to thwart being used for malicious purposes, according to OpenAI. Altman early this year said his company had been "on the wrong side of history" when it came to being open about how its technology works. He later announced that OpenAI will continue to be run as a nonprofit, abandoning a contested plan to convert into a for-profit organization. The structural issue had become a point of contention, with major investors pushing for better returns. That plan faced strong criticism from AI safety activists and co-founder Elon Musk, who sued the company he left in 2018, claiming the proposal violated its founding philosophy. In the revised plan, OpenAI's money-making arm will be open to generate profits but will remain under the nonprofit board's supervision.

OpenAI responds to DeepSeek with two 'open' models

OpenAI is launching GPT-oss-120b and GPT-oss-20b, two open-weight AI models capable of reasoning and complex tasks like coding and online information retrieval. Available on Hugging Face, these models offer text generation capabilities but lack image or video support. While OpenAI discloses the models' numerical parameters, the training data remains proprietary, differing from a fully open-source approach. OpenAI is releasing a pair of open and freely available artificial intelligence models that can mimic the human process of reasoning, months after China's DeepSeek gained global attention with its own open AI software. The two models, called GPT-oss-120b and GPT-oss-20b, will be available on AI software hosting platform Hugging Face and can produce text - but not images or videos-in response to user prompts, OpenAI said on Tuesday. These models can also carry out complex tasks like writing code and looking up information online on a user's behalf, the company said. Crucially, the models are both open-weight systems, similar to Meta Platforms Inc.'s Llama. The term "weight" refers to the parameters in an AI model. OpenAI is disclosing the many numerical values the models picked up and were tweaked with during the training process, allowing developers to better customise them. However, OpenAI is not revealing the data used to train them, falling short of the definition for a truly open source AI model. Despite its name, most of OpenAI's models are closed systems -the kind of software that's controlled by the developer, can't be modified by users, and includes less transparency about its technical underpinnings. Like many of its US rivals, OpenAI has guarded its training data and focused on charging more for its most powerful models to offset the immense cost of development.

OpenAI Unleashes First Open-Weight Models Since GPT-2, Fully Free Under Apache 2.0, With Ability to Run Locally, 128K Context, And Unmatched Customization

In a major leap for the AI industry, OpenAI has officially launched its first set of open-weight models, which marks a pivotal step in introducing transparency and bringing more developer freedom. The two new models, gpt-oss-20b and gpt-oss-120b, are the company's first proper open-weight release after GPT-2 in 2019, and we have been seeing years of closed systems up until now. These two tools are available for free download and can run directly on any hardware with sufficient memory, including Macs with Apple Silicon, denoting a shift in the company's ongoing approach as developers can run AI models locally without the need for servers or APIs. The open-weight language models can be downloaded for free and just went live on GitHub, so developers can immediately gain access to the model weights and inference code under the name GPT-OSS. The models are, however, accompanied by a Complimentary Use Policy so that while the company is focusing on moving away from AI tools being locked behind proprietary walls and giving tech enthusiasts the freedom to run the systems locally without the need for an internet connection or even a server-based API access, it is still maintaining a measured approach. OpenAI states: We aim for our tools to be used safely, responsibly, and democratically, while maximizing your control over how you use them. By using OpenAI gpt-oss-120b and gpt-oss-20b, you agree to comply with all applicable law. So, the legal language the AI company carries is clear: although OpenAI is opening its gates, it is still determined to comply with the laws. But beyond this fine print, this step is monumental because, unlike API-dependent models such as ChatGPT, the new system can run locally on machines with sufficient resources. Developers can build applications without experiencing any latency or dealing with any surveillance issues since it is raw, foundational AI that you have complete control over. If we are to see the technical aspects of these models at a glance, they are rather impressive. The gpt-oss-20b is a 20-billion parameter dense model, while the gpt-oss-120b is a 120-billion parameter MoE model that allows for more computational efficiency and lower inference. The models were additionally trained on a 1.8 trillion-token dataset that includes licensed resources and publicly available data. With a 128K token context window, both models are able to handle complex reasoning and agentic capabilities. The community's reactions to this huge step by OpenAI are largely positive, and many AI researchers are suggesting that OpenAI is focusing on making openness the norm. Noam Brown, a prominent researcher, called this move a step into a multipolar AI ecosystem. Some critics were quick to point out how OpenAI has been late to the open-weight game, with other tech giants such as Meta and Mistral establishing a strong position when it comes to open access. OpenAI's decision nonetheless sends a strong message: it is listening to the developer community and focusing more on transparency and access. One of the most impactful aspects of the release is the Apache 2.0 license under which both models are being offered, which gives developers and other organizations a permissive license to use the models for both commercial and research purposes without any legal hurdles looming. OpenAI taking this leap is vital as it signals a reimagined AI ecosystem where users can customize and deploy local systems without being tied to prohibitive commercial licenses. OpenAI, by moving ahead in this domain, is putting foundational tools in the hands of the public and not behind a paywall, marking a huge shift in the philosophical and strategic direction. This is a win in terms of empowering researchers and developers in the AI space. While it does not turn the company into an open-source company overnight, it is indeed a welcome move, giving the community hope for the beginning of the company's evolving landscape.

OpenAI Returns to Open Source Roots, Releases 120B and 20B AI Models

Both AI models are released under the Apache 2.0 license and can be downloaded from Hugging Face, Ollama, LM Studio, and more. OpenAI has finally released two open-weight AI models called gpt-oss-120b and gpt-oss-20b since the release of GPT-2 in 2019, in a major shift. OpenAI claims that its new open-weight AI models deliver state-of-the-art performance in the open-source arena, outperforming similarly sized models. Both are MoE (Mixture of Experts) reasoning models with 5.1B and 3.6B active parameters. The gpt-oss-120b AI model achieves performance along the lines of o4-mini, and the smaller gpt-oss-20b model is comparable to o3-mini. In its blog post, OpenAI writes, "They were trained using a mix of reinforcement learning and techniques informed by OpenAI's most advanced internal models, including o3 and other frontier systems." Users and developers can run the gpt-oss-120b AI model locally on a single GPU with 80GB VRAM. And the smaller gpt-oss-20b model can run on laptops and smartphones with just 16GB of memory. In addition, these new OpenAI models have a context length of 128k tokens. OpenAI says these models were trained largely on the English dataset, and excel at STEM, coding, and general knowledge. What is interesting is that both open-weight models support agentic workflows, including tool use like web search and Python code execution. It means that you can use these models locally to complete tasks on your computer without requiring an internet connection. Talking about benchmarks, the larger gpt-oss-120b model nearly matches OpenAI's flagship o3 model on Codeforces. In the challenging Humanity's Last Exam benchmark, gpt-oss-120b scores 19% and o3 achieves 24.9% with tools access. Next, in GPQA Diamond, gpt-oss-120b got 80.1% while o3 scored 83.3%. Looking at the benchmarks, OpenAI's open-weight models do look powerful. It would be interesting to see how well gpt-oss-120b and gpt-oss-20b perform against Chinese open-weight AI models such as Qwen and DeepSeek. Both models are released under the Apache 2.0 license and can be downloaded from Hugging Face (120b | 20b).

OpenAI Takes on Rivals Meta, DeepSeek With 2 New Models | PYMNTS.com

By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions. OpenAI's new gpt-oss models come in two sizes: 120 billion and 20 billion parameters, or the statistical relationships learned by a model during training. Generally speaking, the higher the parameter count, the more capable the model. "We believe this is the best and most usable open model in the world," OpenAI CEO Sam Altman said in a post on X. The last time OpenAI released an open model was in 2019, with GPT-2. But GPT-2 was fully open -- making it truly open-source -- albeit done in stages over months. OpenAI didn't say whether gpt-oss will eventually be open-source as well. Gpt-oss is a text-only, open-weights model, meaning the user can use and fine-tune the model but not know how it was trained or what data it was trained on. Without knowing the data that went into the model, companies do not get full transparency, which could add risks to companies in financial services, healthcare and other highly regulated industries. For example, a healthcare company might want to fully vet a model's training process before deploying it on personal patient data. "If only open weights are available, developers ... lack the ability to meaningfully evaluate biases, limitations, and societal impacts," according to the Prompt Engineering and AI Institute. However, OpenAI is granting access under the Apache 2.0 license, which gives the user "perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright" to produce and distribute it. "These open models also lower barriers for emerging markets, resource-constrained sectors, and smaller organizations that may lack the budget or flexibility to adopt proprietary models," according to OpenAI. Major contenders for dominance in the open-source arena are Meta, with its Llama flagship of models but with usage and redistribution restrictions; France's Mistral AI; and Chinese providers such as DeepSeek and Alibaba for most of its Qwen models. Altman called gpt-oss "a big deal," with "strong real-world performance" comparable to o4-mini. (The o, or omni, series comprises OpenAI's reasoning models -- a new family of models after its GPT series.) For example, in the MMLU benchmark that tests how well LLMs perform a broad range of academic and professional tasks, gpt-oss is not far behind OpenAI o3 and o4-mini in performance. OpenAI said the gpt-oss 120-billion parameter model can run on the user's own computer and the smaller model can be run on a smartphone. Typically, AI models especially larger ones, run in the cloud. While releasing an open model means bad actors can use it for evil, Altman said the company thinks "far more good than bad will come from it." This is a departure from 2019, when OpenAI released GPT-2 in stages, fearful that it would used for ill. The rest of OpenAI's models are closed and proprietary. Its rivals have released varying degrees of open models: Google has open-weight, not open-source, models like Gemma. Anthropic doesn't have an open model. Microsoft open-sourced its Phi models. Amazon's models are proprietary. OpenAI Raises New Funding to Hit $300 Billion Valuation

OpenAI releases open-weight reasoning models optimized for running on laptops

SAN FRANCISCO (Reuters) -OpenAI said on Tuesday it has released two open-weight language models that excel in advanced reasoning and are optimized to run on laptops with performance levels similar to its smaller proprietary reasoning models. An open-weight language model's trained parameters or weights are publicly accessible, which can be used by developers to analyze and fine-tune the model for specific tasks without requiring original training data. "One of the things that is unique about open models is that people can run them locally. People can run them behind their own firewall, on their own infrastructure," OpenAI co-founder Greg Brockman said in a press briefing. Open-weight language models are different from open-source models, which provide access to the complete source code, training data and methodologies. Separately, Amazon announced OpenAI's open-weight models are now available on its Bedrock generative AI marketplace in Amazon Web Services. It marks the first time an OpenAI model has been offered on Bedrock, said Atul Deo, Bedrock's director of product. "OpenAI has been developing great models and we believe that these models are going to be great open-source options, or open-weight model options for customers," said Deo, in an interview. He declined to discuss any contractual arrangements between AWS and OpenAI. Amazon shares tumbled last week after the company reported slowing growth in its AWS unit, particularly compared with rivals. The landscape of open-weight and open-source AI models has been highly contested this year. For a time, Meta's Llama models were considered the best, but that changed earlier this year when China's DeepSeek released a powerful and cost-effective reasoning model, while Meta struggled to deliver Llama 4. The two new OpenAI models are the first open models OpenAI has released since GPT-2, which was released in 2019. OpenAI's larger model, gpt-oss-120b, can run on a single GPU, and the second, gpt-oss-20b, is small enough to run directly on a personal computer, the company said. OpenAI said the models have similar performance to its proprietary reasoning models called o3-mini and o4-mini, and especially excel at coding, competition math and health-related queries. The models were trained on a text-only dataset which in addition to general knowledge, focused on science, math and coding knowledge. OpenAI did not release benchmarks comparing the open-weight models to competitors' models such as the DeepSeek-R1 model. Microsoft-backed OpenAI, currently valued at $300 billion, is currently raising up to $40 billion in a new funding round led by Softbank Group. (Reporting by Anna Tong and Greg Bensinger in San Francisco; Editing by Edwina Gibbs and Deepa Babington)