OpenAI releases GPT-6.2 Astra and Sol, Astra 20% cheaper in the API
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, in ChatGPT Work, Codex and the API. Astra is the model for complex professional work and coding, and its API price drops to $8 per million input tokens and $40 per million output tokens, from $10 and $50 for GPT-6 Astra. Sol, the everyday model, keeps its $2 and $10 price while improving on agentic work, coding and writing. The launch comes about a week after Anthropic’s Claude Fable 5.2 and Google’s Gemini 4 Hydrogen.
What’s new
OpenAI positions the two models differently:
- GPT-6.2 Astra is for long, multi-step projects. OpenAI says it builds slides, spreadsheets and documents that follow templates and field conventions, implements and tests complex changes in existing codebases, and has more precise computer use, down to individual parts in CAD and simulation tools. OpenAI calls it “a new quality bar for AI in complex professional work.”
- GPT-6.2 Sol is the general model for everyday work. OpenAI says it is “smarter, more efficient, and lets you complete more tasks on the same paid plan,” and that it can turn a written brief into an interactive browser experience such as a playable game.
In ChatGPT, both models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, 9to5Google reports.
API details
From OpenAI’s model pages for GPT-6.2 Astra and GPT-6.2 Sol:
| GPT-6.2 Astra | GPT-6.2 Sol | |
|---|---|---|
| Model ID | gpt-6.2-astra |
gpt-6.2-sol |
| Input (per 1M tokens) | $8 | $2 |
| Cached input | $0.40 | $0.10 |
| Output | $40 | $10 |
| Context window | 1,050,000 | 1,050,000 |
| Max output | 128,000 | 128,000 |
| Knowledge cutoff | May 18, 2026 | June 1, 2026 |
Both models take text and image input and return text. They work with Chat Completions, Responses and Batch, but not with Realtime, Assistants or fine-tuning. In the Responses API they support web search, file search, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search.
A few pricing rules matter for real workloads:
- Long context: above 272K input tokens, input and cached input rates double and output costs 1.5x for the whole request.
- Batch costs 50% of the standard rate, fast mode 2x.
- Ultrafast costs 3x the standard rate through February 28, 2027, then 6x from March 1, 2027.
- Reasoning effort cannot be set to
none: the API rejects those requests.
Benchmarks
In OpenAI’s own evaluations, as reported by OfficeChai, Astra scores 6.2 points above GPT-6 Astra on its internal GDPval v2 test of professional tasks, and Sol 5.3 points above GPT-6.1 Sol. OpenAI also says Sol comes within half a point of Astra on FrontierCode 2.0. Most other results were published as charts rather than exact numbers, and none have been reproduced independently yet.
What it means for developers
- Switching is a model ID change. If you call
gpt-6-astratoday,gpt-6.2-astrahas the same 1.05M context and 128K output limit at 20% lower input and output prices, and a cached input price cut from $1 to $0.40. OpenAI’s GPT-6 Astra page shows no deprecation notice yet. - Try Sol first for agents. At a quarter of Astra’s price, Sol is the cheaper default for coding agents and tool use; move up to Astra for long codebase changes, document generation or precise computer use.
- Watch the 272K threshold. Long-context requests are billed at higher rates for the entire request, so trimming retrieved context below 272K input tokens can matter more than the base price.
- Other gateways have it too. Vercel’s AI Gateway added both models as
openai/gpt-6.2-astraandopenai/gpt-6.2-solon launch day, with the ultrafast tier available through theservice_tieroption. If you build on the AI SDK, our guide to building AI apps with the Vercel AI framework covers the setup.
How it compares
GPT-6.2 Astra now costs the same as Claude Fable 5.2, which Anthropic priced at $8 per million input tokens and $40 per million output tokens on November 9. OpenAI says both GPT-6.2 models perform roughly on par with Anthropic’s latest models, often at lower cost per task. Google released Gemini 4 Hydrogen in the same week. Benchmark claims from all three companies come from their own testing, so compare them on your own tasks before you switch.