GPT-5.2
GPT-5.2 targets professional knowledge work, reaching 70.9% wins or ties vs experts on GDPval and 52.9% on ARC-AGI-2.
Adds an xhigh reasoning level, better spreadsheets, presentations, vision and long-context use, and an August 2025 knowledge cutoff. 400K context. Priced 1.4x GPT-5.1 at $1.75/$14 per million tokens; Pro $21/$168. Released after a reported "code red" over Gemini 3.
- Date
- Thursday, 11 December 2025
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
- Price
- $1.75 input / $14 output per 1M tokens; Pro $21 / $168
Figures
| Measure | Value | Measured by |
|---|---|---|
| GDPval (wins or ties vs professionals) | 70.9% GPT-5 scored 38.8% | company |
| ARC-AGI-2 | 52.9% GPT-5.1 Thinking 17.6% | company |
| ARC-AGI-1, GPT-5.2 Pro (x-high) | 90.5% at $11.64 per task verified by ARC Prize; o3 preview was 88% at about $4.5k per task a year earlier | independent |
| SWE-Bench Pro (public) | 55.6% GPT-5.1 50.8% | company |
The "code red" memo (reported 2025-12-01 per Simon Willison) is press reporting, not an OpenAI statement. GPT-5.2 was retired from ChatGPT on 2026-06-12 per Wikipedia (not independently confirmed here).
Sources
- openai.com/index/introducing-gpt-5-2/
- simonwillison.net/2025/Dec/11/gpt-52/
- deploymentsafety.openai.com/gpt-5-2/
- developers.openai.com/api/docs/changelog
This record was checked against its sources on 6 October 2026. How we check