o4-mini
by openai
Pricing
Input
$0.44 / 1M tokens
Output
$1.76 / 1M tokens
Overview
o4‑mini is a fast, lightweight reasoning model by OpenAI, launched in April 2025. Built as part of the o-series (with o3), it balances cost-efficiency and strong reasoning—supporting multimodal input, tool use, and high-performance logic at scale.
- Release Date: April 16, 2025
- Multimodal Support: Text + Images (tools enabled)
- Tool Integration: Code interpreter, browser, image generation
- Variants: o4-mini and o4-mini-high (slower but higher quality)
Features
- Fast inference with strong reasoning
- Vision and diagram-based question answering
- Tool use: file upload, code execution, browsing
- High performance on math, science, and QA tasks
- Compatible with ChatGPT "Think" and Team plans
Benchmarks
| Task | o4‑Mini | Claude 3.5 Haiku | DeepSeek V3 |
|---|---|---|---|
| AIME (math) | 99.5%* | 92.0% | 91.0% |
| SWE‑bench (coding) | 68.1% | 64.7% | 63.2% |
| GPQA Diamond (science) | 81.4% | 75.8% | 70.3% |
| VisQA (charts/diagrams) | ✔️ Strong | ✔️ Good | ✔️ Moderate |
*with code interpreter enabled
Pricing
| Type | Price per 1M tokens |
|---|---|
| Input | $1.10 |
| Output | $4.40 |
~10× cheaper than GPT-4 Turbo, with highly competitive performance.
Use Cases
- Scientific and technical reasoning
- Multimodal visual Q&A (diagrams, charts)
- Data analysis with code interpreter
- Real-time coding help and bug fixing
- Educational tutoring and math solvers
Safety and Stability
- Based on OpenAI’s o-series safety architecture
- Moderated for harmful output and misuse
- Instruction-following and adversarial resilience (partial)
- Known vulnerability to "gaslighting" prompt degradation
Limitations
- May degrade under adversarial inputs
- Multimodal still image-only (no audio/video)
- Tool performance varies by plan/tier
License
Commercial usage permitted through OpenAI terms and API plans. Available in ChatGPT Free, Plus, Team, and Enterprise environments.