Introducing the o3 mini AI model
Overview
In the fast-moving world of artificial intelligence, OpenAI has once again pushed the frontier with its new model, o3 mini. Launched in December 2024, this model is presented as the newest and most cost-effective entry in OpenAI’s reasoning-focused series.
Key features of o3 mini
- Strong STEM capabilities: exceptional performance in science, mathematics, and programming.
- Lower cost and lower latency compared with earlier models.
- Developer-focused features: supports function calling, structured outputs, and developer messages.
- Model selector replacement: replaces
o1 minias an option in the ChatGPT model selector.
How it compares to previous models
| Feature | o3 mini | o1 mini |
|---|---|---|
| Response speed | Much faster | Fast |
| STEM capabilities | Advanced | Good |
| Developer features | Full support | Limited |
Availability
According to the original text, o3 mini is available to ChatGPT Plus, Team, and Pro users now, with Enterprise access coming soon. Notably, it’s said to be the first reasoning model accessible to free Persian-language ChatGPT users as well.
For an even easier experience, you can also try o3 mini directly on our all-in-one AI platform at UltraGPT.pro — no setup required.
What you can do with o3 mini
- Get faster, more accurate answers in scientific domains.
- Use advanced programming assistance.
- Access an intelligent assistant with broad domain knowledge.
If you want to try it, the original text suggests checking ChatGPT subscription options to enable o3 mini.
Standout capabilities in science and programming
o3 mini is described as having a special focus on STEM (Science, Technology, Engineering, Mathematics) and demonstrates notable strengths in solving advanced math problems, answering high-level scientific questions, and producing professional code.
Mathematics
- AIME 2024: reported accuracy 83.6% (with high-effort reasoning), outperforming o1 and o1-mini.
- Solved >32% of FrontierMath problems on the first try, including 28% of high-difficulty T3 problems.
Advanced science
- GPQA Diamond (PhD-level science questions): reported accuracy 77.0%.
Programming and software engineering
- Codeforces Elo: reported 2073 (vs. 2061 for o1).
- SWEbench-verified: best performance among OpenAI models with 48.9% accuracy in that benchmark.
- Supports running Python tools for solving complex math problems.
Summary table (from the text):
| Domain | o3 mini (high effort) | o1 |
|---|---|---|
| Math (AIME 2024) | 83.6% | 80.2% |
| Science (GPQA Diamond) | 77.0% | 75.5% |
| Programming (Codeforces Elo) | 2073 | 2061 |
These capabilities make o3 mini a powerful tool for researchers, students, and STEM professionals.
“o3 mini, with its advanced STEM capabilities, is pushing the boundaries of AI and opening new opportunities for researchers and scientists.” — Dr. Ali Mohammadi, AI specialist
Comparative performance on hard benchmarks
- o3 mini reportedly beat o1 and o1-mini across many difficult tests (math, science, programming).
- Human evaluations favored o3 mini over o1-mini in 56% of cases, and the model showed a 39% reduction in major errors on hard, real-world questions.
Key takeaways from the comparisons:
- Most improvement is in STEM fields.
- Increasing the model’s reasoning effort significantly improves performance.
- Human evaluators generally found o3 mini’s answers more accurate and useful.
Speed and efficiency
o3 mini is not only more capable but also faster:
- 24% faster response delivery compared with o1-mini.
- Average response time reduced from 10.16 seconds to 7.7 seconds (o3 mini, medium setting).
Reasoning-effort levels
o3 mini offers three effort levels so users can balance speed and accuracy:
- Low effort: for quick, simple answers.
- Medium effort: balance of speed and accuracy for most tasks.
- High effort: for complex problems requiring deeper reasoning.
Practical benefits of higher speed:
- Less waiting and better UX.
- Faster resolution of complex scientific and coding tasks.
- Higher throughput for bulk workloads and live interactions.
Effects on ChatGPT usage limits
- Message limits for Plus and Team users increased from 50 to 150 messages per day.
- Pro users: unlimited messages.
- Free users: limited access via a “Reason” option in the message composer.
Access table (from the text):
| User type | Daily message limit | Reasoning levels selectable | Integrated search |
|---|---|---|---|
| Plus / Team | 150 | ✓ | ✓ |
| Pro | Unlimited | ✓ | ✓ |
| Free | Limited | ✗ | ✗ |
How to use o3 mini
- Sign in to your ChatGPT account.
- If you have Plus, Team, or Pro, select o3 mini in the model menu.
- Free users: choose the ‘Reason’ option in the composer.
- Pick the desired reasoning effort (if available).
- Start interacting with o3 mini.
Safety and security
The text emphasizes that o3 mini incorporates advanced safety measures, notably a technique called deliberative alignment (i.e., the model is trained to reflect on human-written safety specifications before responding).
Reported safety improvements:
- 95% reduction in producing disallowed content (vs. 80% for earlier models).
- 70% improvement in resistance to jailbreak attempts (vs. 50% previously).
- Extensive internal safety testing, external red-teaming, and large safety evaluations were reportedly performed prior to release.
OpenAI also published a system card for o3 mini that details safety evaluations, identified risks, mitigation steps, and guidance for safe use.
Conclusion
o3 mini is presented as a major step forward: faster, more capable—especially in STEM—and more secure than prior models. It aims to make advanced AI capabilities more broadly accessible, including to Persian-language users without extra tools.