OpenAI unveiled its latest AI reasoning models, o3 and o4-mini, on Wednesday. These models are designed to pause, process, and carefully analyze questions before responding, enhancing their ability to reason through complex tasks.
O3: The Most Advanced Reasoning Model Yet
The o3 model is OpenAI’s most sophisticated reasoning AI to date, surpassing the performance of its previous models in areas such as math, coding, reasoning, science, and visual understanding. According to OpenAI, o3 achieves state-of-the-art performance in tests measuring coding abilities, including a notable 69.1% score on the SWE-bench verified test.
O4-Mini: The Competitive Trade-Off
In contrast, o4-mini offers a more balanced trade-off between price, speed, and performance, making it a versatile option for developers who need a model that is both efficient and effective. This model achieves a performance score of 68.1% on the same SWE-bench test, nearly matching o3’s results.
Developers can choose between the two models depending on their specific needs, with o4-mini presenting a more budget-friendly option while still maintaining high-quality results.
Thinking with Images
One of the most notable features of both o3 and o4-mini is their ability to think with images. Unlike previous reasoning models, these models can now analyze and reason through images, such as whiteboard sketches, diagrams from PDFs, or other visual content uploaded by users. The models can even process blurry or low-quality images, zoom into them, or even rotate them during the “chain-of-thought” phase of reasoning, making these models far more reliable and capable.
Code Execution and Web Search
Another important feature of o3 and o4-mini is the ability to run Python code directly within ChatGPT through the Canvas feature. This allows users to execute code and interact with dynamic data within the conversation. Furthermore, the models can now search the web for current events when prompted, ensuring they stay updated with real-time information.
Pricing and Availability
The new models, including a version of o4-mini-high that takes extra time to generate more reliable answers, are available to OpenAI Pro, Plus, and Team subscribers. For developers, o3 is priced at $10 per million input tokens (roughly 750,000 words) and $40 per million output tokens. o4-mini is offered at a more affordable rate of $1.10 per million input tokens and $4.40 per million output tokens.
In addition to ChatGPT, all three models — o3, o4-mini, and o4-mini-high — will be accessible via OpenAI’s developer-facing endpoints, the Chat Completions API and Responses API, allowing businesses and engineers to integrate these advanced capabilities into their applications.
O3-Pro and GPT-5
Looking ahead, OpenAI plans to release o3-pro, a version of o3 that utilizes additional computing resources to produce even more refined answers, exclusively for ChatGPT Pro subscribers.
Additionally, OpenAI has hinted that these models, o3 and o4-mini, may be among the last stand-alone reasoning models before the arrival of GPT-5, a model expected to unify traditional AI models like GPT-4.1 with reasoning models such as o3 and o4-mini.
Lame . No thanks
So now it can process and reason. It could not do that before lol