GPT-5.1 (Instant and Thinking)
GPT-5.1 makes ChatGPT warmer and more adaptive; the API model adds a default "none" reasoning mode and 24-hour prompt caching.
Instant is warmer and follows instructions better; Thinking spends fewer tokens on easy tasks and persists on hard ones (adaptive reasoning). API adds reasoning effort "none" as default, extended 24-hour cache retention at no extra price, and gpt-5.1-codex models. Personality presets arrived in ChatGPT.
- Date
- Wednesday, 12 November 2025
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Prompt cache retention | up to 24 hours free; KV tensors offloaded to GPU-local storage | company |
Simon Willison identifies November 2025 (GPT-5.1 with Claude Opus 4.5) as when coding agents crossed into daily reliability (2026-09-27 talk).
Sources
- openai.com/index/gpt-5-1/
- simonwillison.net/2025/Nov/13/gpt-51/
- developers.openai.com/api/docs/changelog
- deploymentsafety.openai.com/gpt-5-1/
This record was checked against its sources on 6 October 2026. How we check