OpenAI Introduces o3 And o4-mini: Its Most Advanced Models To Date With Full Tool Access
In Brief
OpenAI has released its latest models, o3 and o4-mini, marking a significant advancement in the capabilities of its reasoning systems.
Artificial intelligence research organization OpenAI, announced the release of its newest models, o3 and o4-mini, which represent an advancement in the capabilities of its reasoning systems.
These models mark the first time ChatGPT can autonomously and intelligently access and integrate all available tools within the platform. This includes the ability to search the web, interpret uploaded documents and data using Python, process visual inputs, and generate images. The models are designed to assess when and how to deploy each tool in order to deliver precise, structured responses—often in under a minute—enabling more effective handling of complex, multi-step queries. This reflects a move toward a more autonomous assistant capable of executing tasks with minimal user direction.
OpenAI describes the o3 model as its most advanced reasoning engine to date. It has set new performance benchmarks across disciplines such as software engineering, mathematics, science, and visual interpretation. Notably, it surpassed previous benchmarks on platforms like Codeforces and SWE-bench, and excelled in academic tests without needing additional customization. The model has been particularly effective in visually intensive tasks, such as analyzing graphs and charts. According to independent evaluators, o3 demonstrated a 20% reduction in major reasoning errors compared to its predecessor, particularly in applied domains such as programming, consulting, and idea generation in scientific and technical contexts. Feedback from early testers emphasized the model’s strong analytical reasoning and its value as a thought partner, particularly in disciplines like biology, engineering, and mathematics.
Meanwhile, the o4-mini model offers a lighter-weight alternative that balances high performance with lower computational costs. Despite its smaller size, it has achieved top scores on competitive assessments like AIME 2024 and 2025, and outperforms the earlier o3-mini in areas beyond STEM, including data analysis and general knowledge tasks. Its efficiency enables broader usage at scale, making it well-suited for high-volume tasks that still require sound reasoning.
Both models have been reviewed by third-party experts, who found notable improvements in their ability to follow instructions and generate helpful, verifiable responses. This is in part due to better integration of web-based sources and improved understanding of context. Compared to earlier versions, o3 and o4-mini also offer a more conversational experience, with enhanced use of memory and context from past interactions to provide more tailored and coherent replies.
Notably, for the first time, these models are able to integrate visual content directly into their reasoning flow. Rather than merely recognizing images, they actively incorporate them into their cognitive process. This capability supports a more advanced form of problem-solving that seamlessly blends visual analysis with text-based reasoning. The models have achieved state-of-the-art results in multimodal tasks, which involve both image and language understanding.
Users can now upload a variety of visual inputs—ranging from textbook diagrams and handwritten notes to whiteboard photos—and the models can interpret them, even under suboptimal conditions like low resolution or reversed orientation. When combined with tool access, the models can modify these images as part of their analysis, adjusting angles or zoom levels as needed to extract relevant information.
This advancement in visual reasoning has expanded the types of tasks the models can handle, delivering high accuracy in areas that were previously challenging for AI systems to address.
OpenAI Rolls Out o3 And o4-mini Models Across ChatGPT Plans, Now Available To Plus, Pro, Team, And Free Users
OpenAI’s latest models, o3 and o4-mini, are equipped with comprehensive access to ChatGPT’s internal tools and can also utilize user-defined tools via application programming interface (API) function calling. These models have been trained not only to execute tasks but to strategically determine which tools to use and when, allowing them to respond to complex prompts with well-structured, relevant outputs. In most cases, they generate these responses in under a minute.
In order to illustrate their capabilities, a reader may consider a scenario where someone asks for a forecast of California’s summer energy use compared to the previous year. The model can independently search the internet for recent utility data, use Python to create a prediction model, generate visuals to illustrate trends, and explain the underlying variables that influence the outcome. What makes this possible is their ability to string together multiple actions, adapting in real-time based on what information they retrieve. If the initial search doesn’t yield enough data, the model can refine its query and continue searching—demonstrating a flexible, iterative reasoning process that mirrors human problem-solving approaches.
This kind of adaptive logic enables o3 and o4-mini to address requests that rely on live data or require a blend of real-time information, in-depth analysis, and multi-format output—something previous models struggled to accomplish without explicit instructions.
Starting today, these models are being rolled out across several ChatGPT plans. Users on Plus, Pro, and Team tiers will see o3, o4-mini, and o4-mini-high in their model selection menus, replacing earlier versions like o1 and o3-mini. Enterprise and Education customers will gain access within a week. For those using the free version of ChatGPT, o4-mini can now be tested by choosing the “Think” option before submitting a prompt. It’s important to note that existing rate limits will remain the same despite the model updates.
Disclaimer
In line with the Trust Project guidelines, please note that the information provided on this page is not intended to be and should not be interpreted as legal, tax, investment, financial, or any other form of advice. It is important to only invest what you can afford to lose and to seek independent financial advice if you have any doubts. For further information, we suggest referring to the terms and conditions as well as the help and support pages provided by the issuer or advertiser. MetaversePost is committed to accurate, unbiased reporting, but market conditions are subject to change without notice.
About The Author
Alisa, a dedicated journalist at the MPost, specializes in cryptocurrency, zero-knowledge proofs, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.
More articles
Alisa, a dedicated journalist at the MPost, specializes in cryptocurrency, zero-knowledge proofs, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.
