On 16 April 2025, OpenAI officially announced the o3 and o4-mini models, representing a new generation of reasoning models capable of autonomously utilising the full suite of ChatGPT tools for the first time, including web search, Python code execution, visual analysis, and image generation.
The o3 model set a new record on the SWE-bench test with a 69.1% score, while the o4-mini delivered a near-identical 68.1% performance at a more affordable price. Both models can "think with images," analysing uploaded sketches or diagrams and performing operations on images during the reasoning process. The o3 and o4-mini outperform their predecessors in complex coding, mathematical, and scientific tasks. Expert evaluations indicate that the o3 makes 20% fewer critical errors than the o1 in challenging real-world tasks, particularly in programming, business consulting, and creative brainstorming.
The models’ pricing is competitive: the o3 costs $10 per million input tokens, while the o4-mini is priced at just $1.10, matching the o3-mini’s rate. OpenAI also introduced the Codex CLI, an open-source coding tool that maximises the capabilities of reasoning models in the terminal. In the coming weeks, OpenAI plans to release the o3-pro model, exclusively for ChatGPT Pro subscribers, as the company gradually integrates GPT and reasoning model capabilities in future developments.
Sources:
1.
Explore OpenAI's latest and most capable models, o3 and o4-mini, designed to think longer before responding and to utilize all tools within ChatGPT, including web browsing, Python execution, and visual reasoning.
2.
Maxwell Zeff
3.
