GPT-4.1, GPT-4.1 mini and GPT-4.1 nano
GPT-4.1 family brings a 1M-token context, 54.6% on SWE-bench Verified and a first nano model, API-only.
Trained for coding, instruction following and long context; 1,047,576-token window and a May 2024 knowledge cutoff. Beats GPT-4o and GPT-4.5 on OpenAI's coding evals at lower cost. It is 26% cheaper than GPT-4o for median queries, and nano at $0.10/$0.40 per million tokens was OpenAI's cheapest model at the time.
- Date
- Monday, 14 April 2025
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
- Price
- nano $0.10/$0.40 per 1M tokens; GPT-4.1 26% cheaper than GPT-4o for median queries
Figures
| Measure | Value | Measured by |
|---|---|---|
| SWE-bench Verified | 54.6% GPT-4o (2024-11-20) 33.2% | company |
| Scale MultiChallenge (instruction following) | 38.3% +10.5 points vs GPT-4o | company |
| Video-MME long, no subtitles | 72.0% +6.7 points vs GPT-4o | company |
| GPT-4.1 nano API price (input / output per 1M tokens) | $0.10 / $0.40 | company |
Available only in the API at launch; reached ChatGPT in 2025-05 (Wikipedia). Announced alongside the deprecation of GPT-4.5 Preview. Retired from ChatGPT on 2026-02-13 (announced 2026-01-29).
Sources
- openai.com/index/gpt-4-1/
- simonwillison.net/2025/Apr/14/gpt-4-1/
- developers.openai.com/api/docs/changelog
This record was checked against its sources on 6 October 2026. How we check