- GPT-6.2 Astra
- GPT-6.2 Sol
Sol was a bigger jump in 3D than I expected. Put it next to what 6.1 Sol made from the same prompt and you can see the difference. 5
Astra made a lovely plane. I found the interface clunky to use. 6
Astra led AA-3D on launch day. On FrontierCode 2.0, OpenAI had Sol within half a point at roughly a fifth of the cost per task, both at Max. Artificial Analysis’s Codex tests were more mixed, with Astra level on its Coding Agent Index and Sol down three points on 6.1 Sol. 1 3
| Model | AA-3D Elo |
|---|---|
| GPT-6.2 Astra | 1,318 |
| Claude Fable 5.2 | 1,271 |
| GPT-6.2 Sol | 1,259 |
Both landed in the API on 17 November. At standard short-context rates, it’s $8 in and $40 out for Astra, or $2 in and $10 out for Sol, per million tokens. OpenAI’s 19 November notes cover Work and Codex. 2 4 7
Sources & references
- OpenAI’s GPT-6.2 Astra and Sol announcement ↗
OpenAI’s launch comparisons include FrontierCode 2.0 at Max effort.
- OpenAI’s model release notes ↗
17 November entry confirms both releases in the API.
- Artificial Analysis on GPT-6.2 Astra and Sol ↗
17 November launch report for AA-3D.
- OpenAI’s GPT-6.2 Sol model documentation ↗
- Our GPT-6.2 Sol seaplane ↗
- Our GPT-6.2 Astra seaplane ↗
- ChatGPT Work and Codex release notes ↗
The 19 November entry covers Work and Codex availability.


