OpenAI has introduced two new AI reasoning models, o3 and o4-mini, which the company says bring significant improvements in logic, coding, science, and image understanding. Launched on April 16, these models are designed to think through questions more carefully before responding and are available now for ChatGPT Pro, Plus, and Team users.
O3 is described as OpenAI’s most capable reasoning model to date, surpassing previous versions in performance across a range of technical benchmarks. It scored 69.1% on the SWE-bench verified test, which measures coding performance, making it the highest-performing OpenAI model in that category. O4-mini and its variant, o4-mini-high, offer a balanced trade-off between cost, speed, and capability, with the standard version scoring 68.1% on the same test.
Both models come with new abilities, including support for ChatGPT’s tools like web browsing, Python code execution, and image generation. A standout feature is their capacity to “think with images.” Users can upload visuals like whiteboard notes or diagrams, which the models analyze during their reasoning process to deliver more informed responses. They can even interpret blurry or low-resolution images, enhancing their versatility.
Developers can also access the models via OpenAI’s Chat Completions and Responses APIs, and pricing has been set competitively. O3 costs $10 per million input tokens and $40 for output, while o4-mini is priced the same as the older o3-mini model at $1.10 and $4.40, respectively.
The launch is seen as a strategic move in the ongoing AI race, with OpenAI aiming to maintain its edge over rivals like Google, Meta, xAI, and Anthropic. Although OpenAI had previously considered holding back o3 in favor of a more advanced unified system, competitive pressures led to its release.
