OpenAI Unveils O1 'Strawberry' Model: A Leap in AI Reasoning and Problem-Solving

OpenAI's expensive new model claims to solve problems like a doctoral candidate

There's also a cheaper, faster o1-mini model that's positioned more towards casual use. All summer long, AI enthusiasts have been looking forward to the next big release from OpenAI, after our first teases about a next-gen project called Strawberry. The model would supposedly offer some big improvements over GPT-4, specifically when it comes to tasks involving programming, mathematics, and logical reasoning. Understandably, we've been excited to check out what it can do, and while we just heard the other day that Strawberry could land within the next two weeks, it's gone ahead and made its debut already, arriving as OpenAI's o1-preview.

OpenAI Just Released Its Long-Awaited "Strawberry" Model

OpenAI has released its long-awaited AI model, reviously code-named "Strawberry." As expected, the new model dubbed "OpenAI o1-preview" -- an entirely new naming convention for the company -- is "designed to spend more time thinking" before responding, pushing the boundaries on the kind of "complex tasks" and "harder problems" it can tackle, according to an update from the company. The model has long been rumored to be a breakthrough in the company's aim to realize artificial general intelligence, the theoretical point at which an AI could outperform a human. The focus is to give the model a sense of "reasoning," enabling it to solve more complex math problems, for instance. And if the company is to be believed, it already has some serious academic chops. OpenAI claims the model "performs similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology." But as its name suggests, the o1-preview is still in a pretty early state and plenty of future updates are to be expected. "As an early model, it doesn't yet have many of the features that make ChatGPT useful, like browsing the web for information and uploading files and images," the company wrote. "For many common cases GPT-4o will be more capable in the near term." As of right now, OpenAI o1 will be available to ChatGPT Plus and Team users. The company is also planning to bring a more lightweight version, dubbed o1-mini, to all free users of ChatGPT, but it has yet to reveal when that will happen. OpenAI says that it designed its latest AI model with safety top of mind. In one of its "hardest jailbreaking tests," the new model scored 84 out of 100, compared to just 22 for its predecessor GPT-4o. The new model "has been trained using a completely new optimization algorithm and a new training dataset specifically tailored for it," OpenAI's research lead Jerry Tworek told The Verge. The company claims o1 could be used "by healthcare researchers to annotate cell sequencing data, by physicists to generate complicated mathematical formulas needed for quantum optics, and by developers in all fields to build and execute multi-step workflows." Thanks to its new "chain of thought" process, it evaluates a number of answers to a query before choosing the best one. And that can take a while, especially when compared to the almost instantaneous answers we get from ChatGPT. In a demo seen by The Verge, the model took 30 seconds to solve a reasoning puzzle involving a princess and prince's age. As for its propensity to "hallucinate" facts, a glaring problem that has historically plagued AI chatbots, OpenAI appeared to be more realistic. "We have noticed that this model hallucinates less," Tworek told The Verge. But "we can't say we solved hallucinations." The company's CEO Sam weighed in on the new model, saying it's "still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it." Does it amount to AGI, one questioner asked the exec. "No," he replied.

With 'Strawberry' AI Model, OpenAI Argues Slower Responses Are Better

OpenAI released a new AI model, dubbed GPT-o1, the culmination of its rumored project Strawberry. The model could eventually be available on the free version of ChatGPT, but for now it's only rolling out to paying ChatGPT Plus and Team users, and in the developer API. Developers will also get access to the more cost-efficient o1-mini variant. How does GPT-o1 differ from its predecessor, GPT-4o, which debuted in May? It takes longer to answer to your questions, which OpenAI argues is a mark of greater intelligence. "We trained these models to spend more time thinking through problems before they respond, much like a person would," OpenAI says. "Through training, they learn to refine their thinking process, try different strategies, and recognize their mistakes." It can explore a train of thought, and therefore processes information differently. While GPT-4o notoriously thinks the word "strawberry" only has two "r"s, GPT-o1 can identify that there are three in a demo video, hence the project code name. GPT-o1 is geared toward researchers, scientists, and developers who work on complex coding and math problems. In internal testing, GPT-o1 solved 83% of questions on the qualifying exam for the International Mathematics Olympiad (IMO), while GPT-4o only solved 13%. The new model also "performs similarly" to PhD students on physics, chemistry, and biology tasks, OpenAI says. CEO Sam Altman says calls GPT-o1 "the beginning of a new paradigm," but admits it's "still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it." In some ways, it's more limited than its predecessor because it cannot browse the web or read uploaded files and images. For those everyday tasks, GPT-4o "will be more capable in the near term," OpenAI says. Though the long-awaited GPT-5 is still under development, OpenAI calls GPT-o1 a critical breakthrough, and possibly a step toward artificial general intelligence (AGI). "I have always believed that you don't need a GPT-6 quality base model to achieve human-level reasoning performance, and that reinforcement learning was the missing ingredient on the path to AGI," says OpenAI researcher Max Schwarzer.

OpenAI Releases its New AI Model, OpenAI o1 (Internally known as Project Strawberry/Q*), a Game-Changer in Reasoning and Competitive Programming!

The model ranks in the 89th percentile on competitive programming questions (!) and correctly solved 83% in a qualifying exam for the International Mathematics Olympiad (IMO) The wait is finally over, and after much speculation, OpenAI has introduced its new o1-preview series of AI models, which can solve complex problems in areas such as science, coding, and math. The models are now available in ChatGPT and the API as part of an early preview, with regular updates and improvements expected. "extremely proud of the team; this was a monumental effort across the entire company.hope you enjoy it!, posted OpenAI chief Sam Altman on X. He even wrapped up his inside joke with AI insider Jimmy Apples by saying, "No more patience, Jimmy," to which Apples replied, "It feels good, Sam. Really good." The o1 series models are trained to spend more time thinking before responding, refining their reasoning process and improving problem-solving capabilities. In initial tests, the next update of the reasoning model performed on par with PhD students on tasks in physics, chemistry, and biology, achieving notable success in math and coding competitions. In a qualifying exam for the International Mathematics Olympiad, the model scored 83%, compared to GPT-4o's 13%. Despite its advanced reasoning abilities, the o1-preview model lacks some of the practical features found in GPT-4, such as browsing the web and file uploading. However, OpenAI emphasises the model's potential for tackling complex tasks, particularly in fields requiring multi-step workflows. As part of the release, OpenAI has implemented a new safety training approach that allows the models to better follow safety rules. In jailbreaking tests, o1-preview outperformed GPT-4o, scoring 84 out of 100 compared to GPT-4o's 22. OpenAI has also bolstered its safety efforts by partnering with AI safety institutes in the U.S. and U.K. Alongside o1-preview, OpenAI has released a smaller, cost-effective model called o1-mini, specifically designed for developers who need advanced coding capabilities without broad world knowledge. O1-mini is 80% cheaper than o1-preview. Starting today, ChatGPT Plus and Team users can manually select o1-preview and o1-mini from the model picker, with rate limits of 30 messages for o1-preview and 50 for o1-mini. API users in the highest usage tier can also begin prototyping, although some features like function calling and streaming are not yet available. OpenAI plans to expand access to o1-mini for ChatGPT Free users and will continue adding new features to the o1 series, including browsing and file uploads. Devin creator, Cognition Labs, worked closely with OpenAI over the last few weeks to evaluate OpenAI o1's reasoning capabilities with Devin. They found that the new series of models represents a significant improvement for agentic systems that deal with code. Altman, too, had hinted a few days earlier in a cryptic post that the company was working on a project known internally as Project Strawberry, also referred to as Q*. "I love summer in the garden," wrote Altman on X, posting the image of a terracotta pot containing a strawberry plant with lush green leaves and small, ripening strawberries. Project Strawberry was said to significantly enhance the reasoning capabilities of OpenAI's AI models. It is pretty clear that o1-preview is exclusively Strawberry. Meanwhile, OpenAI is also in talks to raise funds with an increased valuation of $150 billion. This funding round, led by Thrive Capital, would make Sam Altman's Microsoft-backed company one of the most powerful startups in Silicon Valley.

Strawberry AI is Here: OpenAI Introduces 'o1' Advanced Reasoning Models

OpenAI says o1 models have been trained using a chain-of-thought technique combined with reinforcement learning. After months of anticipation, OpenAI has finally introduced a series of new models called 'o1' that excel at advanced reasoning, which were earlier referred to as Strawberry AI. New models include OpenAI o1, OpenAI o1-preview and OpenAI o1-mini. Preview and mini models are available starting today to paid ChatGPT Plus users. At a later date, OpenAI o1-mini will be available to free ChatGPT users as well. OpenAI says o1 models take some time to think before generating a response, but they can "reason through complex tasks" and solve harder problems in math, science, and coding. In addition, OpenAI says that new reasoning models perform on par with PhD students on challenging science topics. To give you a benchmark, the OpenAI o1 model scored 83% in a rigorous exam like the International Mathematics Olympiad (IMO) whereas GPT-4o could only solve 13% of problems. And in the Codeforces competition, the new o1 model reached the 89th percentile whereas GPT-4o stood at the 11th percentile. In the MMLU benchmark, OpenAI o1 scored 92.3 and on the MATH benchmark, it scored 94.8. OpenAI says in tasks where heavy reasoning is required, o1 closely matches the performance of human experts, which is pretty significant. The o1 models have been trained using a chain-of-thought technique through reinforcement learning. It breaks down the steps into simpler ones and approaches each step through different strategies until it reaches the correct conclusion. By the way, currently, o1 models only support textual input. You can't use the model to browse the web or analyze files and images.

OpenAI unveils a model that can fact-check itself | TechCrunch

ChatGPT maker OpenAI has announced its next major product release: An generative AI model code-named Strawberry, officially called OpenAI o1. Coming soon in both chatbot form and to OpenAI's API, o1 (which is actually a new family of models) avoids some of the reasoning pitfalls that normally trip up generative AI models -- at least according to OpenAI. That's because o1 can effectively fact-check itself by spending more time considering all parts of a command or question. OpenAI says that o1, originating from an internal company project known as Q*, is particularly adept at solving math and programming-related challenges. But what makes the text-only o1 "feel" qualitatively different from other generative AI models is its ability to "think" before responding to queries. When given additional time to "think," o1 can reason through a task holistically -- planning ahead and performing a series of actions over an extended period of time that help it arrive at answers. This makes o1 well-suited for tasks that require synthesizing the results of multiple subtasks, like detecting privileged emails in an attorney's inbox or brainstorming a product marketing strategy. Here's how OpenAI describes it in a blog post: "o1 [was] trained with reinforcement learning to perform complex reasoning. o1 thinks before it answers -- it can produce a long internal chain of thought before responding to the user." TechCrunch wasn't offered the opportunity to test o1 before its debut; we aim to get our hands on it as soon as possible. But according to a person who did have access -- Pablo Arredondo, VP at Thomson Reuters -- o1 is better than OpenAI's previous models (e.g. GPT-4o) at things like analyzing legal briefs and identifying solutions to problems in LSAT logic games. "We saw it tackling more substantive, multi-faceted, analysis," Arredondo told TechCrunch. "Our automated testing also showed gains against a wide range of simple tasks." Now, there is a downside. o1 can be slower than other models, query depending; Arredondo tells us the model can take over ten seconds to answer some questions. (Helpfully, the chatbot version of o1 shows its progress by displaying a label for the current subtask it's performing.) Given the unpredictable nature of generative AI models, o1 likely has other flaws and limitations. We'll no doubt learn about these in time -- and once we get a chance to test the model ourselves. We'd be remiss if we didn't point out that OpenAI is far from the only AI vendor investigating these types of reasoning methods to improve model factuality. Google DeepMind researchers recently published a study showing that, by essentially giving models more compute time and guidance to fulfill requests as they're made, the performance of those models can be significantly improved without any additional tweaks. OpenAI might be first out of the gate with o1. But assuming rivals soon follow suit with comparable models, the company's real test will be making o1 widely available at a reasonable price.

OpenAI's new "reasoning" AI models are here: o1-preview and o1-mini

New o1 model can solve complex tasks due to a new processing step before answering. OpenAI finally unveiled its rumored "Strawberry" AI language model on Thursday, claiming significant improvements in what it calls "reasoning" and problem-solving capabilities over previous large language models (LLMs). Formally named "OpenAI o1," the model family will initially launch in two forms, o1-preview and o1-mini, available today for ChatGPT Plus and API users. OpenAI claims that o1 outperforms its predecessor, GPT-4o, on multiple benchmarks, including competitive programming, mathematics, and "scientific reasoning." However, people who have used the model say it does not yet outclass GPT-4o in every metric. Other users have criticized the delay in receiving a response from the model, owing to the multi-step processing occurring behind the scenes before answering a query. Further Reading In a rare display of public hype-busting, OpenAI product manager Joanne Jang tweeted, "There's a lot of o1 hype on my feed, so I'm worried that it might be setting the wrong expectations. what o1 is: the first reasoning model that shines in really hard tasks, and it'll only get better. (I'm personally psyched about the model's potential & trajectory!) what o1 isn't (yet!): a miracle model that does everything better than previous models. you might be disappointed if this is your expectation for today's launch -- but we're working to get there!" OpenAI reports that o1-preview ranked in the 89th percentile on competitive programming questions from Codeforces. In mathematics, it scored 83 percent on a qualifying exam for the International Mathematics Olympiad, compared to GPT-4o's 13 percent. OpenAI also states that, in what will likely be challenged as people scrutinize the benchmarks and run their own evaluations over time, o1 performs comparably to PhD students on specific tasks in physics, chemistry, and biology. OpenAI attributes these advancements to a new training approach that teaches the model to spend more time "thinking through" problems before responding. This process allows o1 to try different strategies and "recognize" its own mistakes. The smaller o1-mini model is designed specifically for coding tasks and is priced at 80 percent less than o1-preview, offering a more cost-effective option for developers. AI benchmarks are notoriously unreliable and easy to game; however, independent verification and experimentation from users will show the full extent of o1's advancements over time. On top of that, MIT Research showed earlier this year that some of OpenAI's benchmark claims it touted with GPT-4 last year were erroneous or exaggerated. One of the examples of o1's abilities that OpenAI shared is perhaps the least consequential and impressive, but it's the most talked about due to a recurring meme where people ask LLMs to count the number of Rs in the word "strawberry." Due to tokenization, where the LLM processes words in data chunks called tokens, most LLMs are typically blind to character-by-character differences in words. Apparently, o1 has the self-reflective capabilities to figure out how to count the letters and provide an accurate answer without user assistance. So o1 is apparently a mixed bag, and future developments may make it more useful. Wharton Professor Ethan Mollick wrote on X, "Been using GPT-4o1 for the last month. It is fascinating -- it doesn't do everything better but it solves some very hard problems for LLMs. It also points to a lot of future gains." Controversy over "reasoning" terminology It's no secret that some people in tech have issues with anthropomorphizing AI models and using terms like "thinking" or "reasoning" to describe the synthesizing and processing operations that these neural network systems perform. Just after the OpenAI o1 announcement, Hugging Face CEO Clement Delangue wrote, "Once again, an AI system is not 'thinking', it's 'processing', 'running predictions',... just like Google or computers do. Giving the false impression that technology systems are human is just cheap snake oil and marketing to fool you into thinking it's more clever than it is." Reasoning is also somewhat nebulous since it's even difficult to define exactly what the term means in humans. A few hours before the announcement, independent AI researcher Simon Willison tweeted in response to a Bloomberg story about Strawberry, "I still have trouble defining 'reasoning' in terms of LLM capabilities. I'd be interested in finding a prompt which fails on current models but succeeds on strawberry that helps demonstrate the meaning of that term." Reasoning or not, o1-preview currently lacks some features present in earlier models, such as web browsing, image generation, and file uploading. OpenAI plans to add these capabilities in future updates, along with continued development of both the o1 and GPT model series. While OpenAI says the o1-preview and o1-mini models are rolling out today, neither model is available in our ChatGPT Plus interface yet, so we have not used them. We'll report our impressions on how this model differs from other LLMs we have previously covered. This is a breaking news story that will be updated.

OpenAI Releases o1 Model, With Reasoning Capabilities

OpenAI is releasing a new artificial intelligence model known internally as "Strawberry" that can perform some human-like reasoning tasks, as it looks to stay at the top of a crowded market of rivals. The new model, called o1, is built to "spend more time thinking" before responding to user queries, the company said in a blog post Thursday. With the model, OpenAI's tools should be able to solve multi-step problems, including complicated math and coding questions. A preview version of the model is available to ChatGPT Plus and Team users today. "As an early model, it doesn't yet have many of the features that make ChatGPT useful, like browsing the web for information and uploading files and images," the company said. "But for complex reasoning tasks this is a significant advancement and represents a new level of AI capability. Given this, we are resetting the counter back to 1 and naming this series OpenAI o1." The model's release comes as OpenAI is looking to raise billions in funding and faces heightened competition in the race to develop ever more sophisticated artificial intelligence systems. OpenAI isn't the only company working on such capabilities; competitors Anthropic and Google have also touted "reasoning" skills with their advanced AI models. Bloomberg previously reported the company could release the new model as soon as this week.

OpenAI launches o1: Next-gen AI model that thinks, codes like a PhD

Codenamed Strawberry, the new model spends more time thinking through problems before they respond, much like a person would. ChatGPT developer OpenAI has announced the launch of a new series of AI reasoning models to solve hard problems on September 12. Codenamed Strawberry, the new AI models are officially called OpenAI o1. Interestingly, OpenAI has trained these models to spend more time thinking through problems before responding, much like a person would. According to OpenAI, the models can reason through complex tasks and solve harder problems than previous models in science, coding, and math.

OpenAI's new AI model can perform some human-like reasoning tasks including solving complicated math problems

OpenAI is releasing a new artificial intelligence model known internally as "Strawberry" that can perform some human-like reasoning tasks, as it looks to stay at the top of a crowded market of rivals. The new model, called o1, is designed to spend more time computing the answer before responding to user queries, the company said in a blog post Thursday. With the model, OpenAI's tools should be able to solve multi-step problems, including complicated math and coding questions. "As an early model, it doesn't yet have many of the features that make ChatGPT useful, like browsing the web for information and uploading files and images," the company said. "But for complex reasoning tasks this is a significant advancement and represents a new level of AI capability. Given this, we are resetting the counter back to 1 and naming this series OpenAI o1." A preview version of the model will be available through OpenAI's popular chatbot, ChatGPT, to paid Plus and Team users on Thursday. Bloomberg previously reported the company could release the new model as soon as this week. The model's release comes as San Francisco-based OpenAI is looking to raise billions in funding and faces heightened competition in the race to develop ever more sophisticated artificial intelligence systems. OpenAI isn't the only company working on such capabilities; competitors Anthropic and Google have also touted "reasoning" skills with their advanced AI models. In its blog post, OpenAI gave examples of the AI model's responses to questions on topics including coding, English, and math, and asked it to solve a simple crossword puzzle. In a series of posts on X, Noam Brown, a research scientist at OpenAI, said the company is releasing the model in preview now in part to get a sense for how people use it, and where it needs to be improved.

OpenAI Strawberry is here - it's called o1-preview and it might be the most human ChatGPT ever

Fast answers aren't always the best, which might be the key takeaway from the arrival of OpenAI Strawberry - now called o1-preview - a new ChatGPT reasoning model that takes longer to give you what might be vastly better answers. OpenAI announced the preview release on Thursday in a blog post, saying that it will arrive in ChatGPT and the generative AI company's API. I can confirm that the o1-preview and a faster, cheaper model o1-mini are both live in our ChatGPT Plus account. The new models will not yet appear in the free ChatGPT accounts, though. Strawberry has been eagerly anticipated because of its possible human-like-thinking capabilities. In the weeks before this announcement, OpenAI CEO Sam Altman has teased us with numerous cheeky fruit references, but has also made it clear in recent months that generative AI was set to make a significant leap forward. In the blog post, OpenAI explained, "We trained these models to spend more time thinking through problems before they respond, much like a person would. Through training, they learn to refine their thinking process, try different strategies, and recognize their mistakes." OpenAI claims this more powerful o1-preview has performed "similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology." And that's key here. o1-preview is a generative model that might have the greatest application in academia, not for helping you write an engaging prom-posal. One example given in a video accompanying the blog is gene sequencing. In it, a scientist notes that while humans can't keep track of everything in gene sequencing, an AI can. The scientist refers to the new model as "chat with reasoning" and shows how when she types in a question, there's a moment where o1-preview says "Thinking." The value of it is that it keeps her from rabbit-holing into the wrong part of gene theory. However, o1-preview is not a replacement for ChatGPT-4o, which is barely a month old. The new model isn't searching the web or capable of ingesting files and images. Though, that will likely show up at some point. If biology and math are not your thing, the lighter and slightly more agile o1-mini might be for you, and is also live in ChatGPT Plus now. It's particularly adept at coding. You can try out the new models in ChatGPT Plus ($20 a month) by logging in and then selecting the model drop-down menu. You'll see o1-preview, and o1-mini have been added to the list as of this story publishing. It will be interesting to see what people do with the model formerly known as Strawberry and if people grow impatient while waiting for an o1-preview to deliver answers. Perhaps they'll feel better if they can see that the extra reasoning is resulting in more useful answers for some really big problems.

Meet o1: OpenAI's advanced reasoning 'Strawberry' model

Initially dubbed "Project Strawberry," the model was released to the public as "o1" on Thursday. As expected, o1 is able to "think" before answering a prompt, which differs from previous large language models that use predictive-text to form their queries. In a demo shared on the OpenAI site, the video shows a user asking o1 in ChatGPT to "write a grammatictlly [sic] correct sentence without using any letters more than once." The model then uses its new reasoning skills to weigh options and evaluate the sentence. According to the demo, after 39 seconds, o1's cheeky response to the query is "go fix my bed." OpenAI's o1 model is available through the developer AI and ChatGPT Plus users in preview. When you go to the ChatGPT app, you can select o1 in the dropdown menu.