OpenAI Unveils o3, o4-mini: Smarter, Faster Reasoning AI

Curated by Ayaan Zaveri
Apr 19, 2025

Summary

OpenAI's new models, o3 and o4-mini, are designed to think more deeply and improve the quality of AI responses by combining advanced reasoning with full tool access github.blog+1.
Both models can agentically use tools like web search, Python coding, and image understanding, enabling more independent, multi-step problems to be solved effectively github.blog+1.
They incorporate "thinking with images", allowing the models to interpret, manipulate, and reason through visual information such as diagrams, sketches, and charts github.blog+1.
o3 excels in coding, math, science, and visual perception, setting state-of-the-art (SOTA) benchmarks, while o4-mini is a cost-efficient, high-performance model for fast reasoning, particularly in math and coding tasks github.blog+1.
Pricing varies, with o3 costing $10.00 per 1,000 tokens for input and $40.00 for output, while o4-mini is more affordable at $1.10 and $4.40 respectively zdnet.com.
They are safety-stressed, with refreshed safety training and rigorous safety evaluations, including safety prompts related to biological, cybersecurity, and AI risk factors github.blog+1.

12 references