OpenAI Unveils Advanced AI Reasoning Models: o3 and o4-mini

OpenAI launches a pair of AI reasoning models, o3 and o4-mini | TechCrunch

OpenAI announced on Thursday the launch of o3 and o4-mini, new AI reasoning models designed to pause and work through questions before responding. The company calls o3 its most advanced reasoning model ever, outperforming the company's previous models on tests measuring math, coding, reasoning, science, and visual understanding capabilities. Meanwhile, o4-mini offers what OpenAI says is a competitive trade-off between price, speed, and performance -- three factors developers often consider when choosing an AI model to power their applications. Unlike previous reasoning models, o3 and o4-mini can generate responses using tools in ChatGPT such as web browsing, Python code execution, image processing, and image generation. Starting today, the models, plus a variant of o4-mini called o4-mini-high that spends more time crafting answers to improve its reliability, are available for subscribers to OpenAI's Pro, Plus, and Team plans. The new models are part of OpenAI's effort to beat out Google, Meta, xAI, Anthropic, and DeepSeek in the cutthroat global AI race. While OpenAI was first to release an AI reasoning model, o1, competitors quickly followed with versions of their own that match or exceed the performance of OpenAI's lineup. In fact, reasoning models have begun to dominate the field as AI labs look to eke more performance out of their systems. O3 nearly wasn't released in ChatGPT. OpenAI CEO Sam Altman signaled in February that the company intended to devote more resources to a sophisticated alternative that incorporated o3's technology. But competitive pressure seemingly spurred OpenAI to reverse course in the end. OpenAI says that o3 achieves state-of-the-art performance on benchmarks including Codeforces, SWE-bench (without building a custom model-specific scaffold), and MMMU. OpenAI claims that o3 and o4-mini are its first models that can "think with images." In practice, users can upload images to ChatGPT, such as whiteboard sketches or diagrams from PDFs, and the models will analyze the images during their "chain-of-thought" phase before answering. Thanks to this newfound ability, o3 and o4-mini can understand blurry and low-quality images and can perform tasks such as zooming or rotating images as they reason. Beyond image processing capabilities, o3 and o4-mini can run and execute Python code directly in your browser via ChatGPT's Canvas feature, and search the web when asked about current events. In addition to ChatGPT, all three models -- o3, o4-mini, and o4-mini-high -- will be available via OpenAI's developer-facing endpoints, the Chat Completions API and Responses API, allowing engineers to build applications with the company's models at usage-based rates. In the coming weeks, OpenAI says it plans to release o3-pro, a version of o3 that uses more computing resources to produce its answers, exclusively for ChatGPT Pro subscribers. OpenAI CEO Sam Altman has indicated o3 and o4-mini may be its last standalone AI reasoning models in ChatGPT before GPT-5, a model that the company has said will unify traditional models like GPT-4.1 with its reasoning models.

OpenAI's GPT-o3 Reasoning Model Is Ready for Prime Time

Katelyn is a writer with CNET covering social media, AI and online services. She graduated from the University of North Carolina at Chapel Hill with a degree in media and journalism. You can often find her with a novel and an iced coffee during her time off. OpenAI o3, the developer's most advanced generative AI reasoning model yet, is here. The new family of models promises strong performance in coding, math, science and visual understanding, the company said in a press release on Wednesday. The new model is available for paying ChatGPT Plus, Pro and Team users. Free users can also try out the new reasoning capability by selecting "Think" from the drop down menu before entering a prompt. Education and enterprise users will have access in one week. You'll still have your usual rate limits when you use the new model. OpenAI touts o3 as a smart AI model with the ability to reason (meaning it can recursively check its answers before giving you a final output) and to independently use all of ChatGPTs' tools, including web browsing, Python, image understanding and image generation. OpenAI says this helps solve complex multi-step problems. Along with o3, OpenAI also announced o4-mini, a faster and cheaper reasoning model that can take on math, coding and visual tasks. Another model, o4-mini-high, is meant for more complex tasks that require more reasoning time. OpenAI says these models will be replacing older models, including o1, o3-mini and o3-mini-high. OpenAI says both o3 and o4-mini can "think with images," which is marketing lingo meaning the model can understand the images you upload to the chat. The new models should be able to "see" and incorporate the images into their reasoning chain. OpenAI can train on the information you share unless you opt out, so make sure you aren't uploading sensitive or personal information. For example, if you're asking ChatGPT o3 a chemistry question, you should be able to upload formulas scribbled on a whiteboard for the AI to use in its answer. CNET hasn't tested this out yet, but the ability does seem in line with its native image generator's ability to create legible, semi-accurate text in AI images. You can also crop and rotate the images, which is a new image editing ability in ChatGPT. For coders, OpenAI is releasing an AI agent that runs locally on a user's terminal. Called Codex CLI, it's a lightweight and open-source agent that can take advantage of o3 and o4-mini, with GPT-4.1 support coming later. For those requiring even more computing power, OpenAI is expected to release o3-pro in a few weeks. Until then, Pro users can still access the existing o1-pro model. With Google and OpenAI trading blows for AI dominance, OpenAI has been on a surge with new feature releases. On Monday, the company launched GPT-4.1, a new family of generative AI models with a 1 million token context window -- essentially, the amount of information the model can process. While it's a major improvement over 4.0, Google's Gemini 1.5 Pro has a larger, 2 million token context window. Not to be outdone, OpenAI introduced an update that greatly improved the memory of ChatGPT earlier this month. Despite Google's continued efforts to take the AI crown, ChatGPT is still the default AI system for most people. ChatGPT is currently the leader in AI with a 60% market share with 400 million weekly users, according to software firm Neontri. Even though Google dominates in online search, Gemini only has a 13.5% market share. Considering the AI market is expected to be worth $1.01 trillion by 2031, according to Statista, companies that claim dominance early have the biggest chance of taking the largest part of that share. OpenAI is also eyeing new ways to source data -- and compete in another arena with tech billionaires and AI enthusiasts Elon Musk and Mark Zuckerberg. The AI company is considering building its own social media feed, The Verge reported this week. Meta and X/Twitter are both able to use its troves of user data to train their AI models. As OpenAI continues to build more advanced models like o3 and o4-mini, it will need an increasingly large, steady stream of human-generated content to refine the models' outputs. A new social media platform could be one way for the company to find that data.

OpenAI's upgraded o3 model can use images when reasoning

Jay Peters is a news editor covering technology, gaming, and more. He joined The Verge in 2019 after nearly two years at Techmeme. OpenAI is releasing two new AI reasoning models today: o3, which the company calls its "most powerful reasoning model," and o4-mini, which is a smaller and faster model that "achieves remarkable performance for its size and cost," according to a blog post. The company also says that o3 and o4-mini will be able to "think" with images, meaning they will "integrate images directly into their chain of thought." That could be useful if you show the models things like sketches or whiteboards. OpenAI says that the models will also be able to adjust images by zooming in on them or rotating the image "as part of their reasoning process." In addition, OpenAI is announcing that its reasoning models will be able to use all ChatGPT tools, including things like web browsing and image generation. The tools will be available today for ChatGPT Plus, Pro, and Team users in o3, o4-mini, and o4-mini-high, and will come to o3-pro in "a few weeks." (o1, o3-mini, and o3-mini-high will be phased out from those tiers.)

OpenAI just dropped new o3 and o4-mini reasoning AI models - and a surprise agent

Following the recent launch of a new family of GPT-4.1 models, OpenAI released o3 and o4-mini on Wednesday, its latest addition to the existing line of reasoning models. The o3 model, previewed in December, is OpenAI's most advanced reasoning model to date, while o4-mini is a smaller, cheaper, and faster model. Simply put, reasoning models are trained to "think before they speak," which results in a longer time to process the prompt but higher-quality responses. As a result, like older models, o3 and o4-mini show strong performance in coding, math, and science tasks. However, they also have an important new addition: Visual understanding. Also: How to use ChatGPT: A beginner's guide to the most popular AI chatbot OpenAI o3 and o4-mini are OpenAI's first models to "think with images." OpenAI explains that this means the models don't just see an image; they can actually use the visual information in their reasoning process. Users can also now upload images that are low quality or blurry, and the model will still be able to understand them. Another major first is that o3 and o4-mini can independently use all ChatGPT tools, including web browsing, Python, image understanding, and image generation, to better resolve complex, multi-step problems. OpenAI says this ability allows the new models to take "real steps toward acting independently." Also: The top 20 AI tools of 2025 - and the #1 thing to remember when you use them A recent report from The Information claimed the new models would synthesize information from different fields and subject expertise and then use that knowledge to suggest new, innovative experiments. Insider sources who have tested the model said these experiments would encompass many complex topics, such as nuclear fission or pathogen detection, according to the report. OpenAI has yet to confirm this. OpenAI o3 and o4-mini are available today to subscribers, including ChatGPT Plus, Pro, and Team users. The models will appear in the model picker as o3, o4-mini, and o4-mini-high, replacing o1, o3-mini, and o3-mini-high. Pro users will be able to access o3-pro in a few weeks, but until then, they will still have access to o1-pro. The models are also available for developers via the API. Also: ChatGPT just made it easy to find and edit all the AI images you've ever generated To ease concerns about model safety, OpenAI shared that both of the new releases were stress-tested under its safety program and evaluated under its updated Preparedness Framework. OpenAI has also launched Codex CLI, an open-source coding agent that runs locally in users' terminals. It is meant to provide users with a simple and clear way to connect AI models, including o3 and o4-mini (with support for GPT-4.1 coming soon), to their own code and tasks running on their computer. Also: OpenAI to launch AI models that can think up their own experiments, says report OpenAI announced the launch of a $1 million initiative well. It's meant to support early projects by awarding grants in $25K increments via API credits.

OpenAI to launch AI models that can think up their own experiments, says report

Dubbed o3 and o4-mini, the new models could arrive as early as this week - at a rather staggering cost. Following the launch of a brand new family of GPT-4.1 models, a new report from The Information claims OpenAI will soon release two new models capable of performing a task typically reserved for humans -- coming up with scientific experiments. Also: It's a private cloud revival: Why Kubernetes and cloud-native tech are essential in the AI age The new models, dubbed o3 and o4-mini, would be additions to OpenAI's existing line of reasoning models. Simply put, these reasoning models are trained to "think before they speak," which results in a longer time to process the prompt but higher quality responses. These types of models are especially useful for STEM tasks, which would be the focus of o3 and o4-mini. According to the report, the new models will synthesize information from different fields and subject expertise and then use that knowledge to suggest new, innovative experiments. According to insider sources who have tested the model and are cited in the report, these experiments would encompass many complex topics, such as nuclear fission or pathogen detection. These capabilities are particularly noteworthy because the high level of reasoning necessary to create viable experiments has typically been left to humans. Furthermore, combining insights from different specialties, a method used by inventors like Nikolas Tesla, is especially daunting. The breadth of what these models can be used for goes beyond scientific applications; businesses, think tanks and other organizations can equally benefit. Questions regarding accuracy and hallucinations remain to be explored. Also: My 5 favorite AI apps on Android right now - and how I use them The advanced assistance will come at a cost. OpenAI will reportedly charge $20,000 monthly for this "doctorate-level AI," which is 100 times more expensive than ChatGPT Pro, OpenAI's most expensive subscription tier right now, which retails at $200 per month. The exact date of when the model will be released is to be determined, but the report says it could be as early as this week. OpenAI originally previewed the o3 and o3-mini models on its last day of 'shipmas' in December. At the time, OpenAI said that the technology would not be available to the general public yet, with only early testers having access. Then, in January, the company launched o3-mini in ChatGPT and the API. Benchmark evaluations highlighted o3-mini's proficiency in STEM-related tasks.

OpenAI Releases New Reasoning Models for Coding and Visual Tasks

OpenAI is rolling out a pair of new artificial intelligence models that mimic the process of human reasoning to field more complicated coding questions and visual tasks, the latest in a flurry of releases from the company to fend off renewed competition in the US and China. The first model, called o3, spends more time computing an answer before responding to user prompts, with the goal of solving more complex multistep problems related to science, math and coding, the company said Wednesday. OpenAI is also releasing a model called o4-mini that performs well in similar categories but is meant to be more compact and nimble. Both are now available to OpenAI's paid users.

OpenAI says newest AI model can 'think with images,' understanding diagrams and sketches

Sam Altman Co-founder and CEO of OpenAI speaks during the Italian Tech Week 2024 at OGR Officine Grandi Riparazioni on September 25, 2024 in Turin, Italy. OpenAI has released its latest artificial intelligence model, which it said is capable of "thinking with images," meaning it can understand and analyze a user's sketches and diagrams, even if they're low quality. The main new reasoning model from OpenAI is called o3, and the company simultaneously released a smaller model dubbed o4-mini. The rollout follows the September debut of OpenAI's first reasoning model, o1, which focused on solving complex problems and deliberating over its answers in multiple steps. With o3, users can upload whiteboards, sketches and other images and have the AI analyze and discuss them. The models can also rotate, zoom and use other image-editing tools. Since introducing its viral ChatGPT chatbot in late 2022, OpenAI has been rapidly upgrading its models to go well beyond text into images, voice and videos. The company is racing to stay ahead in generative AI, where it faces heated competition from rivals including Google, Anthropic and Elon Musk's xAI. "For the first time, our reasoning models can independently use all ChatGPT tools -- web browsing, Python, image understanding, and image generation," OpenAI wrote. "This helps them solve complex, multi-step problems more effectively and take real steps toward acting independently." The company, valued at $300 billion in a funding round last month, said that o3 and o4-mini are its first AI models that can "think with images." That means "they don't just see an image, they can integrate visual information directly into the reasoning chain," according to OpenAI. Last month, OpenAI released a native image-generation feature that went viral online for its ability to produce Studio Ghibli-style anime images. OpenAI said its o3 model is especially tuned for math, coding, science and understanding images, while o4-mini operates faster and at a lower cost. Both models were available starting Wednesday to ChatGPT Plus, Pro and Team customers. OpenAI's user community has long joked about its strange or confusing AI model names. CEO Sam Altman joined the banter this week, writing in a post on X, "how about we fix our model naming by this summer and everyone gets a few more months to make fun of us (which we very much deserve) until then?" The company also said that both models had been "stress-tested under our most rigorous safety program to date" and linked to its "Preparedness framework" updated earlier this week. OpenAI has come under fire recently for changes to its safety precautions and processes. It said this week that it reserved the right to "change its safety requirements if 'another frontier AI developer releases a high-risk system without comparable safeguards.'" In changing its policies this week, OpenAI wrote it would no longer require safety tests for certain fine-tuned models. The company also avoided releasing a "model card" -- or a report containing information about safety tests done before a model is released -- for its GPT-4.1 model. In February, OpenAI launched AI agent tool Deep Research weeks before publishing its system card. OpenAI didn't immediately respond to a request for comment.

OpenAI just launched its smartest AI yet that can think with images -- here's how to try it

OpenAI just released two updated AI models -- o3 and o4-mini -- for ChatGPT Plus, Pro and Team users. Essentially two new, bigger and better brains, these models are said to be the smartest ones yet because they can tackle more advanced queries, understand the blurriest images, and solve problems like never before. This release comes just a few days after OpenAI announced that ChatGPT is getting a major upgrade to its memory features, aimed at making conversations even more personal, seamless and context-aware. With ChatGPT retiring GPT-4 at the end of this month, the release of these new models underscore OpenAI's broader push to make ChatGPT feel less like a one-off assistant and more like a long-term, adaptable tool that evolves with its users. These models are the most advanced yet, capable of interpreting both text and images, including lower-quality visuals such as handwritten notes and blurry sketches. Users can upload diagrams or whiteboard photos, and the models will incorporate them into their responses. The models also support real-time image manipulation, such as rotating or zooming, as part of the problem-solving process. For the first time, the models can independently use all of ChatGPT's tools, including the browser, Python code interpreter, image generation and image analysis. This means the AI can decide which tools to use based on the task given, potentially making it more effective for research, coding, and visual content creation. As part of this launch OpenAI is also unveiling Codex CLI, an open-source coding agent that runs locally in a terminal window. It's designed to work with these new models and will soon support GPT-4.1. To encourage developers to test and build with these tools, OpenAI is offering $1 million in API credits, distributed in $25,000 increments. The newly released o3 and o4-mini models are now available to ChatGPT Plus subscribers, with developers able to access them via the OpenAI API. A more advanced o3-pro model is expected to arrive in the coming weeks. In the meantime, users on the Pro plan can continue using the existing o1-pro model. These updates come at a time when OpenAI is no longer held back by limited computing power -- a shift that could mark a major leap forward for AI. In a recent interview with Business Insider, CEO Sam Altman revealed that OpenAI is no longer "compute constrained," meaning the company now has access to the kind of massive processing power needed to build more sophisticated models. With this boost it looks likely that OpenAI can accelerate development, roll out more powerful versions of ChatGPT, and create models capable of handling far more complex tasks. In short, the brakes are officially off. This newfound capacity also signals OpenAI's broader ambition to make its models more flexible, intelligent, and autonomous, particularly for users who rely on AI for research, content creation and coding. As these tools evolve, so does the potential for AI to move beyond assistant-level support and become a true creative and analytical collaborator.

OpenAI Releases Smarter AI Models

OpenAI today announced the release of new o3 and o4-mini AI models, which the company says are its smartest models to date. o3 and o4-mini can use and combine all ChatGPT tools, including web searches, analyzing files and data with Python, reasoning about visual input, and generating images. The models have been trained to reason about when and how to use different tools to provide detailed and thorough information, and to better understand and answer multi-faceted questions. They are able to "think" with images, blending visual and textual reasoning. o3 excels at coding, math, science, and visual perception, and OpenAI says that it's ideal for complex queries where an answer is not immediately obvious, and it is adept at visual tasks like analyzing charts and graphics. o4-mini is a smaller model that is optimized for quick and cost-efficient reasoning, and it is ideal for math, coding, and visual tasks. It supports higher usage limits than o3. Both of the new models are designed to be more natural and conversational, offering more useful and verifiable responses. o3, o4-mini, and o4-mini-high are available to ChatGPT Plus, Pro, and Team users as of today, replacing o1, o3-mini, and o3-mini-high. Free users are able to try o4-mini by using the "Think" option in the composer when submitting a query. Rate limits remain the same.

OpenAI launches o3 and o4-mini, AI models that 'think with images' and use tools autonomously

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI launched two groundbreaking AI models today that can reason with images and use tools independently, representing what experts call a step change in artificial intelligence capabilities. The San Francisco-based company introduced o3 and o4-mini, the latest in its "o-series" of reasoning models, which it claims are its most intelligent and capable models to date. These systems can integrate images directly into their reasoning process, search the web, run code, analyze files, and even generate images within a single task flow. "There are some models that feel like a qualitative step into the future. GPT-4 was one of those. Today is also going to be one of those days," said Greg Brockman, OpenAI's president, during a press conference announcing the release. "These are the first models where top scientists tell us they produce legitimately good and useful novel ideas." How OpenAI's new models 'think with images' to transform visual problem-solving The most striking feature of these new models is their ability to "think with images" -- not just see them, but manipulate and reason about them as part of their problem-solving process. "They don't just see an image -- they think with it," OpenAI said in a statement sent to VentureBeat. "This unlocks a new class of problem-solving that blends visual and textual reasoning." During a demonstration at the press conference, a researcher showed how o3 could analyze a physics poster from a decade-old internship, navigate its complex diagrams independently, and even identify that the final result wasn't present in the poster itself. "It must have just read, you know, at least like 10 different papers in a few seconds for me," Brandon McKenzie, a researcher at OpenAI working on multimodal reasoning, said during the demo. He estimated the task would have taken him "many days just for me to even like, onboard myself, back to my project, and then a few days more probably, to actually search through the literature." The ability for AI to manipulate images in its reasoning process -- zooming in on details, rotating diagrams, or cropping unnecessary elements -- represents a novel approach that industry analysts say could revolutionize fields from scientific research to education. Beyond AI models: How o3 and o4-mini function as complete AI systems with advanced tool integration OpenAI executives emphasized that these releases represent more than just improved models -- they're complete AI systems that can independently use and chain together multiple tools when solving problems. "We've trained them to use tools through reinforcement learning -- teaching them not just how to use tools, but to reason about when to use them," the company explained in its release. Greg Brockman highlighted the models' extensive tool use capabilities: "They actually use these tools in their chain of thought as they're trying to solve a hard problem. For example, we've seen o3 use like 600 tool calls in a row trying to solve a really hard task." This capability allows the models to perform complex, multi-step workflows without constant human direction. For instance, if asked about future energy usage patterns in California, the AI can search the web for utility data, write Python code to analyze it, generate visualizations, and produce a comprehensive report -- all as a single fluid process. OpenAI surges ahead of competitors with record-breaking performance on key AI benchmarks OpenAI claims o3 sets new state-of-the-art benchmarks across key measures of AI capability, including Codeforces, SWE-bench, and MMMU. In evaluations by external experts, o3 reportedly makes 20 percent fewer major errors than its predecessor on difficult, real-world tasks. The smaller o4-mini model is optimized for speed and cost efficiency while maintaining strong reasoning capabilities. On the AIME 2025 mathematics competition, o4-mini scored 99.5 percent when given access to a Python interpreter. "I really do believe that with this suite of models, o3 and o4-mini, we're going to see more advances," Mark Chen, OpenAI's head of research, said during the press conference. The timing of this release is significant, coming just two days after OpenAI unveiled its GPT-4.1 model, which excels at coding tasks. The rapid succession of announcements signals an acceleration in the competitive AI landscape, where OpenAI faces increasing pressure from Google's Gemini models, Anthropic's Claude, and Elon Musk's xAI. Last month, OpenAI closed what amounts to the largest private tech funding round in history, raising $40 billion at a $300 billion valuation. The company is also reportedly considering building its own social network, potentially to compete with Elon Musk's X platform and to secure a proprietary source of training data. How OpenAI's new models transform software engineering with unprecedented code navigation abilities One area where the new models particularly excel is software engineering. Brockman noted during the press conference that o3 is "actually better than I am at navigating through our OpenAI code base, which is really useful." As part of the announcement, OpenAI also introduced Codex CLI, a lightweight coding agent that runs directly in a user's terminal. The open-source tool allows developers to leverage the models' reasoning capabilities for coding tasks, with support for screenshots and sketches. "We're also sharing a new experiment: Codex CLI, a lightweight coding agent you can run from your terminal," the company announced. "You can get the benefits of multimodal reasoning from the command line by passing screenshots or low fidelity sketches to the model, combined with access to your code locally." To encourage adoption, OpenAI is launching a $1 million initiative to support projects using Codex CLI and OpenAI models, with grants available in increments of $25,000 in API credits. Inside OpenAI's enhanced safety protocols: How the company protects against AI misuse OpenAI reports conducting extensive safety testing on the new models, particularly focused on their ability to refuse harmful requests. The company's safety measures include completely rebuilding their safety training data and developing system-level mitigations to flag dangerous prompts. "We stress tested both models with our most rigorous safety program to date," the company stated, noting that both o3 and o4-mini remain below OpenAI's "High" threshold for potential risks in biological, cybersecurity, and AI self-improvement capabilities. During the press conference, OpenAI researchers Wenda and Ananya presented detailed benchmark results, noting that the new models underwent over 10 times the training compute of previous versions to achieve their capabilities. When and how you can access o3 and o4-mini: Deployment timeline and commercial strategy The new models are immediately available to ChatGPT Plus, Pro, and Team users, with Enterprise and Education customers gaining access next week. Free users can sample o4-mini by selecting "Think" in the composer before submitting queries. Developers can access both models via OpenAI's Chat Completions API and Responses API, though some organizations will need verification to access them. The release represents a significant commercial opportunity for OpenAI, as the models appear both more capable and more cost-efficient than their predecessors. "For example, on the 2025 AIME math competition, the cost-performance frontier for o3 strictly improves over o1, and similarly, o4-mini's frontier strictly improves over o3-mini," the company stated. The future of AI: How OpenAI is bridging reasoning and conversation for next-generation systems Industry analysts view these releases as part of a broader convergence in AI capabilities, with models increasingly combining specialized reasoning with natural conversation abilities and tool use. "Today's updates reflect the direction our models are heading in: we're converging the specialized reasoning capabilities of the o-series with more of the natural conversational abilities and tool use of the GPT-series," OpenAI noted in its release. Ethan Mollick, associate professor at the Wharton School who studies AI adoption, described o3 as "a very strong model, but still a jagged one" in a social media post after the announcement. As competition in the AI space continues to intensify, with Google, Anthropic, and others releasing increasingly powerful models, OpenAI's dual focus on both reasoning capabilities and practical tool use suggests a strategy aimed at maintaining its leadership position by delivering both intelligence and utility. With o3 and o4-mini, OpenAI has crossed a threshold where machines begin to perceive images the way humans do -- manipulating visual information as an integral part of their thinking process rather than merely analyzing what they see. This shift from passive recognition to active visual reasoning may ultimately prove more significant than any benchmark score, representing the moment when AI began to truly see the world through thinking eyes.

OpenAI unveils technology that can 'reason' with images

Unlike early versions of its ChatGPT chatbot, these reasoning systems spend a significant amount of time "thinking" about a question before answering, rather than providing an instant response. The systems are part of a wider effort to build AI that can reason through complex tasks. Companies like Google, Meta and DeepSeek, a Chinese startup, are developing similar technologies.In September, OpenAI introduced artificial intelligence technology that could "reason" through tasks involving math, coding and science. Now, this technology can tackle similar tasks that involve images, including sketches, posters, diagrams and graphs. On Wednesday, the company unveiled two new versions of its reasoning technology called OpenAI o3 and OpenAI o4-mini. Each can handle tasks that involve both images and text. These systems can "manipulate, crop and transform images in service of the task you want to do," said Mark Chen, head of research at OpenAI, in announcing the new system during an internet livestream. OpenAI also said these systems could generate images, search the web and use other digital tools. Unlike early versions of its ChatGPT chatbot, these reasoning systems spend a significant amount of time "thinking" about a question before answering, rather than providing an instant response. The systems are part of a wider effort to build AI that can reason through complex tasks. Companies like Google, Meta and DeepSeek, a Chinese startup, are developing similar technologies. The goal is to build systems that can solve a problem through a series of steps, each one building on the last, similar to how humans reason. These technologies can be particularly useful to computer programmers who use AI systems to write code. The reasoning systems are based on a technology called large language models, or LLMs. To build reasoning systems, companies put LLMs through an additional process called reinforcement learning. During this process, a system learns behavior through extensive trial and error. By working through various math problems, for instance, it can learn which methods lead to the right answer and which do not. If it repeats this process with a large number of problems, it can identify patterns. OpenAI's latest systems have learned to handle problems that involve both images and text. Experts point out that reasoning systems do not necessarily reason like a human. And like other AI technologies, they can get things wrong and make stuff up -- a phenomenon called hallucination. OpenAI also unveiled a new tool, Codex CLI, that is designed to further facilitate computer programming tasks that involve systems like o3 and o4-mini. Called an AI agent, it provides ways of using these AI systems in tandem with existing code stored on a programmer's personal machine. The company said it was open sourcing this tool, meaning it is freely sharing its underlying technology with programmers and businesses, allowing them to modify and build on the technology. OpenAI said that, beginning Wednesday, these new systems would be available to anyone who subscribed to ChatGPT Plus, a $20-a-month service, or ChatGPT Pro, a $200-a-month service that provides access to all of the company's latest tools. (The New York Times has sued OpenAI and its partner, Microsoft, for copyright infringement of news content related to AI systems. Both companies have denied the claims).

OpenAI Releases o3 and o4-mini, Says o3 Can 'Generate Novel Hypotheses'

The o3 model delivers state-of-the-art performance in coding, math, science, and visual tasks. In December 2024, OpenAI announced o3, its most advanced reasoning AI model, and said the model will be released after proper safety testing. Finally, the frontier AI lab has launched the full o3 AI model after a gap of four months. Along with that, OpenAI has also released the next-generation o4-mini (and o4-mini-high) reasoning model. In these four months, OpenAI has improved the o3 model even further and says o3 is the "most powerful reasoning model" developed by the company. Both o3 and o4-mini models can use multiple agentic tools inside ChatGPT, including web search, Python tools, and more. The reasoning models can finally analyze images as well. Both o3 and o4-mini are trained to pick the right tools, depending on the task. OpenAI says o3 sets a new benchmark in coding, math, science, and visual tasks such as analyzing images, charts, and graphics. Early testers say that o3 can "generate and critically evaluate novel hypotheses -- particularly within biology, math, and engineering contexts." On the other hand, the new o4-mini is a smaller model, designed for speed and cost-efficiency. It excels in math, coding, and visual tasks. In fact, the smaller o4-mini model achieves 99.5% on AIME 2025 when given access to a Python interpreter. As for benchmarks, both models have nearly saturated AIME 2024 and 2025. However, on GPQA Diamond, o3 achieves 83.3 and o4-mini gets 81.4. On Humanity's Last Exam, o3 (without tools) scores 20.32 and with tools, gets 24.9. Finally, on SWE-Bench Verified, the o3 model scores 69.1%, even higher than Google's Gemini 2.5 Pro (63.8%). On multimodal benchmarks, both models are pretty competitive and achieve high accuracy in MMMU, MathVista, and CharXiv-Reasoning. Lastly, OpenAI also released Codex, a new command-line agentic tool, somewhat similar to Anthropic's Claude Code. You can run it from your terminal and take advantage of multimodal reasoning using o3 and o4-mini. As for availability, o3 and o4-mini are rolling out to ChatGPT Plus, Pro, and Team users, starting today. The two new models will replace o1, o3-mini, and o3-mini-high. OpenAI says ChatGPT Enterprise and Edu users will get access in one week. Thankfully, o4-mini is also coming to free-tier ChatGPT users, which can be accessed through the 'Think' button. OpenAI has also assured that o3-pro is coming in a few weeks with support for all tools. Meanwhile, ChatGPT Pro users can continue to use the o1-pro model. In case you missed the 2024 announcement, OpenAI's o3 reasoning model was the first to crack the ARC-AGI benchmark, scoring an impressive 87.5% on the ARC-AGI Semi-Private Evaluation set in a high-compute configuration. François Chollet, the creator of ARC-AGI, noted in a blog post: This is not merely incremental improvement, but a genuine breakthrough, marking a qualitative shift in AI capabilities compared to the prior limitations of LLMs. o3 is a system capable of adapting to tasks it has never encountered before, arguably approaching human-level performance in the ARC-AGI domain. However, it was also revealed that o3 had been trained on 75% of the ARC-AGI Public Training set, raising questions about how much of o3's performance relied on generalized intelligence or benchmark-specific tuning. Nevertheless, a recent report from The Information reveals that o3 can blend information from multiple fields like Nikola Tesla. It can come up with novel scientific ideas and experiments in areas like nuclear fusion and pathogen detection. In fact, OpenAI reportedly believes that its capabilities are powerful enough to justify a $20,000 per month pricing tier and calls it a "PhD-level AI."

OpenAI set to release AI models for innovative problem-solving - The Information By Investing.com

Investing.com -- OpenAI is reportedly gearing up to release new artificial intelligence (AI) models, potentially as early as this week, that can generate innovative ideas across various fields, ranging from nuclear fusion to pathogen detection, according to a report from The Information. The models, known as o3 and o4-mini, have been tested by a select group of individuals who were not authorized to publicly discuss the matter. These AI models are designed to bridge the gap between different disciplines to suggest novel types of experiments. If these models perform as initial testers indicate, they could present new solutions for AI customers, including the design and discovery of new materials or drugs. This could potentially draw interest from Fortune 500 companies, including oil and gas firms and commercial drug developers, as well as research lab scientists. OpenAI, the creator of the ChatGPT, introduced reasoning models in September. These models are tailored to excel in problems with solutions that can be objectively verified, such as mathematical theorems. They also perform better when given more time to process answers. This shift towards reasoning models occurred last year as traditional methods of improving AI started to plateau. The impending release of these models signifies a significant advancement in AI's ability to brainstorm and generate new ideas, a task previously thought to be primarily human-led. This development indicates a significant step in AI's capabilities, challenging the belief that humans would always be the primary source of new discoveries.

OpenAI announces o3 and o4-mini AI reasoning models: Here's what they can do

These models are designed to pause and carefully work through questions before providing responses. OpenAI has unveiled two new AI reasoning models: o3 and o4-mini. These models are designed to pause and carefully work through questions before providing responses. According to OpenAI, these new AI models are the smartest models they've released to date. The o3 model is OpenAI's most powerful reasoning model, outperforming previous models in key areas like math, coding, reasoning, science, and visual understanding. It achieves top-tier performance in coding tests, scoring 69.1% on the SWE-bench verified test, a significant improvement compared to earlier models. Meanwhile, o4-mini offers a balanced mix of price, speed, and performance. "OpenAI o4-mini is a smaller model optimized for fast, cost-efficient reasoning -- it achieves remarkable performance for its size and cost, particularly in math, coding, and visual tasks," the AI giant said in a blogpost. Also read: OpenAI working on its own social media platform to rival Elon Musk's X, says report One of the key features of both o3 and o4-mini is their ability to generate responses by combining every tool in ChatGPT, such as web browsing, Python code execution, and image processing. These models are also the first from OpenAI to think with images. "People can upload a photo of a whiteboard, a textbook diagram, or a hand-drawn sketch, and the model can interpret it -- even if the image is blurry, reversed, or low quality. With tool use, the models can manipulate images on the fly -- rotating, zooming, or transforming them as part of their reasoning process," the company added. Also read: OpenAI introduces new GPT-4.1 AI models: Here's everything you should know Starting today, ChatGPT Plus, Pro, and Team users will see o3, o4-mini, and o4-mini-high in the model selector, replacing the previous options: o1, o3-mini, and o3-mini-high. ChatGPT Enterprise and Edu users will gain access in one week. Free users can try o4-mini by selecting 'Think' in the composer before submitting their query. In the future, OpenAI plans to release o3-pro. With the new models, OpenAI continues to push the boundaries of AI.