Simon Willison shipped LLM 0.38 same-day support, and Astra lands at 80% of GPT-6 Astra pricing.
OpenAI introduced two models on November 17, 2026: GPT-6.2 Astra for the hardest projects and GPT-6.2 Sol for everyday work. Both ship through the API as gpt-6.2-astra and gpt-6.2-sol, and Simon Willison released LLM 0.38 with same-day support for both.
Astra costs $8 per million input tokens and $40 per million output tokens, 80% of the GPT-6 Astra rate and the same as Claude Fable 5.2. Cached input drops to $0.40 per million tokens from $1 for the older model. Sol holds at $2 input and $10 output, half the price of GPT-5.6 Sol.
Cheaper cached tokens matter most for long agentic sessions that resend large contexts, and LLM 0.38 support lets teams test both models within hours of launch. Willison’s Codex CLI runs at Ultra effort show neither model morphing a map scene into a mech as cleanly as Claude Fable 5.2 or Gemini 4 Carbon.
Early reception is mixed. Astra 6.2 trails Claude Fable 5.2 on several benchmarks, and reviewers including Theo Browne flagged a confusing communication style during pre-release testing. Willison reports the model manages subagents well on long tasks, then asks operators for odd internal implementation details.
What matters
- OpenAI released GPT-6.2 Astra for hard projects and GPT-6.2 Sol for everyday work.
- Astra cached input costs $0.40 per million tokens, a discount that favors long agentic runs.
- Mixed early reception and benchmark gaps versus Fable 5.2 will shape enterprise adoption.
Why it matters
Mixed early reception and benchmark gaps versus Fable 5.2 will shape enterprise adoption.
This GenAI News article was prepared in original wording using reporting and materials published by Simon Willison’s Weblog. Source reference: https://simonwillison.net/2026/Nov/17/gpt62-astra-and-sol/.
Drafted by the GenAI News review pipeline.
