SS
About Me
Frontier AI Paper BriefingsPokebowlClinical Trial EnrollerLittle Human Names
DisclaimersPrivacy PolicyTerms of Use
Privacy Policy·Terms of Use·Disclaimers

© 2026 Silvia Seceleanu

← Back to Explorer
Models·OpenAI·Apr 2025

33. Introducing o3 and o4-mini

Reasoning models get tools

Product Announcement
Summary

Released o3 and o4-mini, next-generation reasoning models with tool use capabilities. o3 achieved state-of-the-art on ARC-AGI and complex benchmarks, while o4-mini delivered strong reasoning at lower cost — best model on AIME 2024/2025.

Key Concepts

o3: most capable reasoning model at launch, with tools woven into the reasoning loop

The most capable reasoning model at release, with native tool use integrated into the reasoning loop. Could call tools mid-thought, use results to update its reasoning, and iterate.

o4-mini: best-in-class math performance (AIME 2024/2025) at a fraction of the cost

Optimized for speed and cost. Despite being smaller, achieved best-in-class performance on AIME 2024 and 2025 mathematical competitions. Outperformed o3-mini on non-STEM tasks as well.

The model thinks with tools — interleaving reasoning steps with code execution and web search

The critical advance was that reasoning and tool use were integrated — the model doesn't just think, then use tools. It thinks with tools, interleaving reasoning steps with code execution and web search.

Connections

33. Introducing o3 a…Apr 202530. 12 Days of OpenA…Dec 202434. GPT-5 / Codex CL…Aug 2025Influenced byInfluences
Influenced by
30. 12 Days of OpenAI: o3, Sora, and More
Dec 2024
Influences
34. GPT-5 / Codex CLI / Research Agent
Aug 2025