Become a member of our daily and weekly newsletters for the latest updates and exclusive content about leading AI coverage. Leather
Openi Today launched two groundbreaking AI models that can reason with images and can use independently of the tools, which represents which experts call a step change in artificial intelligence options.
The company established in San Francisco introduced O3 And o4-miniThe newest in his “O-series” of reasoning models, which she believes are the most intelligent and capable models to date. These systems can integrate images directly into their reasoning process, search the internet, perform code, analyze files and even generate images within a single stream of task.
“There are some models that feel like a qualitative step in the future. GPT-4 was one of that. Today it will also be one of those days,” said Greg Brockman, president of OpenAi, during a press conference that announced the release. “These are the first models where top scientists tell us that they produce good and useful new ideas legitimate.”
How OpenAi’s new models ‘think with images’ to transform visual problem solving
The most striking function of these new models is their ability to ”Think with images“-Not just seeing, but also manipulating and reasoning about them as part of their problem-solving process.
“They don’t just see an image – they think,” Openai said in a statement sent to Venturebeat. “This unlocks a new class of problem solving that combines visually and textual reasoning.”
During a demonstration at the press conference, a researcher showed how O3 could analyze a physics poster of a decade old internship, navigating independently of his complex diagrams and even identified that the end result was not present in the poster itself.
“It must have just read, you know, at least 10 different articles for me in a few seconds,” said Brandon McKenzie, a researcher at OpenAi who works on multimodal reasoning, during the demo. He estimated that the task would have been given him “many days alone for me to even keep on board, back to my project, and then a few days more likely, to actually look through literature.”
The ability of AI to manipulate images in his reasoning process – zoom in on details, rotating diagrams or the cropping of unnecessary elements – is a new approach that industrial analysts say they can give a revolution in fields of scientific research into education.
I had early access, O3 is an impressive model, seems very capable. Some nice examples:
– Ethan Mollick (@Emollick) April 16, 2025
1) A business case that I use in my class
2) Some SVGs make (images only made by code)
3) Writing a limited story of two intersecting gyres
4) Hard Science Fiction Space Battle. pic.twitter.com/tk4ppvnot
OpenAI leaders emphasized that these releases represent more than just improved models -these are complete AI systems that can use multiple tools independently and chain in solving problems.
“We trained them to use tools through reinforcement learning – they not only teach how to use tools, but to reason when they have to use them,” the company explained in the release.
Greg Brockman emphasized the extensive uses of the models: “They actually use these tools in their thoughts, because they try to solve a difficult problem. For example, we have seen O3 use such as 600 tool calls in a row trying to resolve a really hard task.”
With this possibility, the models can perform complex workflows with multiple steps without constant human direction. If asked for future energy consumption patterns in California, for example, the AI can search the internet for utility – data, write python code to analyze it, generate visualizations and produce an extensive report – all as a single liquid process.
OpenAI is responsible for competitors with record-breaking performance on key AI benchmarks
OpenAI Claims O3 sets new state-of-the-art benchmarks in important measures of AI capacities, including Codeforces” SweatAnd Mmmu. In evaluations of external experts, O3 reportedly makes 20 percent less large mistakes than its predecessor on difficult, real-world tasks.
The smaller O4-mini model is optimized for speed and cost efficiency while retaining strong reasoning options. On the AIME 2025 Maths competitionO4-Mini scored 99.5 percent when access was given to a Python-Tolk.
“I really believe that with this series of models, O3 and O4-Mini, we will see more progress,” said Mark Chen, OpenAi’s main investigation, during the press conference.
The timing of this release is considerable, only two days after OpenAi was revealed GPT-4.1 ModelIt excels in coding tasks. The rapid sequence of announcements indicates an acceleration in the competitive AI landscape, where OpenAi is confronted with increasing pressure from Google’s Gemini modelsAnthropics Clamberand Elon Musk’s Xai.
Last month, OpenAi closed, which amounts to the largest private technical financing round in history, Pick up $ 40 billion With a rating of $ 300 billion. The company is said to also be considering building its own social networkPossibly to compete with the X platform of Elon Musk and to secure its own source of training data.
O3 and O4-Mini are super good at coding, so we release a new product, Codex Cli, to make them easier to use.
This is a coding agent that runs on your computer. It is completely open source and available today; We expect it to improve soon.
– Sam Altman (@Sama) April 16, 2025
How OpenAi’s new models of software -engineering transform with unprecedented code -navigation skills
An area where the new models mainly excel is software engineering. Brockman noted during the press conference that O3 “is actually better than I am in navigating through our OpenAI code base, which is really handy.”
As part of the announcement, OpenAi has also introduced Codex CliA lightweight coding agent that is carried out directly in the terminal of a user. With the Open Source Tool, developers can use the reasoning options of the models for coding tasks, with support for screenshots and sketches.
“We also share a new experiment: Codex CLI, a lightweight coding agent that you can perform from your terminal,” the company announced. “You can get the benefits of multimodal reasoning of the command rule by passing on screenshots or low reliability sketches to the model, combined with access to your code locally.”
To encourage adoption, OpenAi launches a $ 1 million initiative To support projects using Codex Cli and OpenAI models, with subsidies available in steps of $ 25,000 to API credits.
In the Enhanced Safety Protocols of OpenAI: how the company protects against AI abuse
OpenAI reports that performing extensive safety tests on the new models, in particular aimed at their ability to refuse harmful requests. The company’s safety measures include the complete rebuilding of their safety educational data and the development of mitigations at system level to mark dangerous instructions.
“We emphasize both models with our most rigorous safety program so far,” the company stated, and noted that both O3 and O4-Mini remain under the “high” threshold of OpenAI for potential risks in organic, cyber security and AI self-improvement.
During the press conference, OpenAI researchers Wenda and Ananya presented detailed benchmark results and noted that the new models underwent more than 10 times the training of previous versions to achieve their possibilities.
When and how you have access to O3 and O4-Mini: Implementment Timeline and Commercial Strategy
The new models are immediately available for Chatgpt Plus” ProAnd Team users, with Venture And Education Customers who will have access next week. Free users can taste O4-Mini by selecting “Think” in the composer before submitting queries.
Developers have access to both models via OpenAi’s Chat -volaties API And Answers APIAlthough some organizations need verification to gain access to them.
The release is an important commercial opportunity for OpenAI, because the models seem to be both more capable and more cost -efficient than their predecessors. “With the 2025 Aime Math Competition, for example, the cost-performance boundary for O3 is strictly improving compared to O1, and in the same way the limit of O4-Mini is strictly improving in relation to O3-Mini,” the company said.
The Future of AI: How OpenAi bridges the reasoning and a conversation for the next generation systems bridges
Industrial analysts consider these releases as part of a broader convergence in AI options, in which models are increasingly combining specialized reasoning with natural conversation options and tool use.
“Today’s updates reflect the direction in which our models go: we converge the specialized reasoning options of the O-series with more of the natural conversation skills and the use of the GPT series,” Openai noted in the release.
Ethan Mollick, associate professor at the Wharton School who studies AI -Adoption, described O3 as “a very strong model, but still a crazy” in a social media post after the announcement.
While the competition in the AI space continues to intensify, with Google, anthropic and others who release more and more powerful models, OpenAI’s Double focus suggests both reasoning options and practical tool use a strategy aimed at maintaining the leadership position by providing both intelligence and use.
With O3 and O4-Mini, OpenAi has crossed a threshold where machines start to observe images like people manipulating visual information as an integral part of their thinking process instead of just analyzing what they see. This shift from passive recognition to active visual reasoning can ultimately be more important than any benchmark score, which represents the moment when AI really started to see the world through thinking eyes.
Source link
