Model Details
o3-mini
Quick answer: o3-mini is OpenAI's January 2025 compact reasoning model, scoring 97.9% on MATH, 86.9% on MMLU, and 15.0% on SimpleQA. It brings o3-class mathematical and scientific reasoning at a lower cost than full o3, with three effort levels: low, medium, and high.
At a Glance
Where o3-mini leads
- 97.9% MATH — near-perfect competition mathematics performance
- 86.9% MMLU — strong general knowledge for a reasoning-focused model
- Cost-efficient reasoning: lower price than o3 and o4-mini
- Three reasoning effort levels (low/medium/high) for cost-quality tradeoff
- Strong STEM performance: coding, mathematics, science
Where it lags
- 15.0% SimpleQA — reasoning models sacrifice factual recall for reasoning depth
- Text-only: no vision capability
- Higher latency than GPT-4o-mini due to reasoning steps
- Superseded by o4-mini for most reasoning tasks with added vision support
Best for: Cost-sensitive reasoning tasks; STEM problem solving; API integration requiring reasoning at low cost; teams evaluating o-series models.
What o3-mini Is
o3-mini (released January 31, 2025) is OpenAI's small reasoning model in the o3 family. Like o1 and o3, it uses extended chain-of-thought reasoning before responding — trading latency for accuracy on complex reasoning tasks.
Its 97.9% MATH score represents near-ceiling performance on competition mathematics. The model was designed for developers who need o3-class reasoning quality but are cost-constrained by full o3 pricing. Three effort levels (low/medium/high) allow tuning the latency-cost-accuracy tradeoff.
o4-mini (April 2025) superseded o3-mini with better scores across all benchmarks, added vision capability, and similar pricing. For new deployments, o4-mini is the recommended choice unless o3-mini has been specifically validated.
Specifications
| Field | Value |
|---|---|
| Organization | OpenAI |
| Context window | 200,000 tokens |
| License | Proprietary (API only) |
| Release date | January 31, 2025 |
| Knowledge cutoff | September 2024 |
| Modality | Text only |
| Reasoning effort | Low / Medium / High |
Pricing
| Tier | Price |
|---|---|
| Input | $1.10 / 1M tokens |
| Output | $4.40 / 1M tokens |
Pricing from the OpenAI API pricing page. Cached input tokens are discounted.
Context Window
o3-mini supports a 200,000-token context window — approximately 150 pages of text.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| MATH | 97.9% | Benchgen evaluation | 2025-07 |
| MMLU | 86.9% | Benchgen evaluation | 2025-07 |
| SimpleQA | 15.0% | Benchgen evaluation | 2025-07 |
o3-mini vs Alternatives
| Model | MATH | MMLU | Context | Price (in/out) |
|---|---|---|---|---|
| o3-mini | 97.9% | 86.9% | 200K | $1.10/$4.40 |
| o4-mini | — | — | 200K | $1.10/$4.40 |
| o1-preview | 85.5% | 90.8% | 128K | — |
| o4-mini-high | — | — | 200K | $1.10/$4.40 |
o3-mini vs o4-mini: o4-mini achieves better benchmark scores with added multimodal capability at identical pricing. For new deployments, o4-mini is recommended.
Related Models & Benchmarks
- Compare: o4-mini · o4-mini-high · o1-preview
- Benchmarks: MATH · MMLU · SimpleQA
Frequently Asked Questions
What is o3-mini?
o3-mini is OpenAI's January 2025 compact reasoning model, scoring 97.9% MATH and 86.9% MMLU. It brings o3-class reasoning at lower cost with three effort levels.What is o3-mini's context window?
o3-mini supports a 200,000-token context window.Should I use o3-mini or o4-mini?
o4-mini for new deployments: better benchmarks, multimodal capability, and same pricing ($1.10/$4.40 per 1M tokens). o3-mini is appropriate for existing integrations.What is the difference between o3-mini low, medium, and high?
Effort levels control how much reasoning the model performs. Low: faster, cheaper, lower accuracy. High: slower, more expensive, highest accuracy. Medium is a balanced default.Specs from OpenAI's o3-mini launch (January 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.