Everything we know: OpenAI's o3 model

Has OpenAI o3 been released?

As of January 31, 2025, OpenAI o3-mini has been officially released, marking a major step forward in cost-effective reasoning. o3-mini is now available in ChatGPT and the API for select developers in API usage tiers 3-5, and for ChatGPT Plus, Team, and Pro users starting today. Enterprise access is expected in February.

The full o3 model has not yet been released, but OpenAI continues safety testing and refinements. More information on its launch is expected soon.

OpenAI o3-mini: Cost-Effective and Powerful Reasoning

OpenAI o3-mini is a fast, cost-efficient model optimized for STEM reasoning, particularly excelling in science, math, and coding. It builds upon OpenAI o1-mini, offering superior accuracy, speed, and reasoning capabilities while maintaining a low cost.

o3-mini introduces several key features:

• Function Calling, Structured Outputs, and Developer Messages – making it production-ready from launch.
• Three Reasoning Effort Modes (Low, Medium, High) – allowing developers to optimize performance based on complexity and latency needs.
• Higher Rate Limits – ChatGPT Plus and Team users now have 150 messages per day (up from 50 messages with o1-mini).
• Search Integration – o3-mini can now retrieve up-to-date answers with linked sources.
• First Reasoning Model for Free Users – available in ChatGPT’s free plan under the “Reason” option.

Limitations: o3-mini does not support vision capabilities, so OpenAI o1 remains the preferred choice for visual tasks.

What to Expect from OpenAI o3

The upcoming OpenAI o3 model is expected to surpass o1 in reasoning, problem-solving, and general knowledge capabilities, further extending OpenAI’s leadership in AI research. While o3-mini provides an optimized, cost-effective option for technical domains, the full o3 model will likely push AI intelligence to new heights.

Performance of OpenAI o3-mini

OpenAI o3-mini has undergone extensive testing and outperforms o1-mini across multiple benchmarks:

• Mathematics: o3-mini matches or exceeds o1’s performance in challenging exams like AIME (American Invitational Mathematics Examination).
• Science: On PhD-level science questions (GPQA Diamond), o3-mini achieves 77% accuracy at high reasoning effort.
• Coding: o3-mini excels in Codeforces competitive programming, achieving 2073 Elo at high reasoning effort.
• Software Engineering: o3-mini is the best OpenAI model on SWE-bench Verified tasks.
• Human Preference Testing: Testers preferred o3-mini’s responses 56% of the time over o1-mini, with a 39% reduction in major errors.

Speed and Efficiency

o3-mini delivers faster responses than its predecessor:

• 24% faster response times than o1-mini.
• 2.5 seconds faster time-to-first-token latency improvement.

Safety and Responsible AI

OpenAI trained o3-mini using deliberative alignment, ensuring strong safety performance and reduced jailbreak vulnerabilities. External red-team testing and safety evaluations confirm o3-mini significantly outperforms GPT-4o in safety compliance.

What’s Next?

The release of OpenAI o3-mini demonstrates OpenAI’s commitment to cost-effective, high-performance AI. With expanded access, improved reasoning, and superior efficiency, o3-mini sets the stage for the eventual release of OpenAI o3, expected to push AI capabilities even further.

Stay tuned for updates on OpenAI o3’s full release, which is expected to continue OpenAI’s tradition of groundbreaking advancements in reasoning, coding, and problem-solving.

About PromptLayer

PromptLayer is a prompt management system that helps you iterate on prompts faster — further speeding up the development cycle! Use their prompt CMS to update a prompt, run evaluations, and deploy it to production in minutes. Check them out here. 🍰