GPT-5.4 mini and GPT-5.4 nano
GPT-5.4 mini runs over twice as fast as GPT-5 mini and nears GPT-5.4 on SWE-Bench Pro and OSWorld; nano targets classification and subagents.
Small models for high-volume work. Mini scores 54.4% on SWE-Bench Pro and 72.1% on OSWorld-Verified at xhigh effort; nano is the cheapest. Input prices rose 3x (mini) and 4x (nano) versus GPT-5 mini and nano.
- Date
- Tuesday, 17 March 2026
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
- Price
- mini $0.75/$4.50; nano $0.20/$1.25 per 1M tokens
Figures
| Measure | Value | Measured by |
|---|---|---|
| SWE-Bench Pro (mini / nano, xhigh) | 54.4% / 52.4% GPT-5.4 57.7%, GPT-5 mini 45.7% | company |
| OSWorld-Verified (mini / nano, xhigh) | 72.1% / 39.0% GPT-5.4 75.0% | company |
| API price GPT-5.4 mini / nano (input per 1M tokens) | $0.75 / $0.20 output $4.50 / $1.25 | company |
Sources
- openai.com/index/introducing-gpt-5-4-mini-and-nano/
- simonwillison.net/2026/Mar/17/mini-and-nano/
- developers.openai.com/api/docs/changelog
This record was checked against its sources on 6 October 2026. How we check