OpenAI on Tuesday, November 17, released GPT-6.2 Astra, its frontier model for demanding professional and scientific work, and GPT-6.2 Sol, its cheaper everyday model, in ChatGPT Work, Codex and the API. Astra’s API price falls 20 percent to $8 per million input tokens and $40 per million output tokens, while Sol keeps GPT-6.1 Sol’s $2 and $10. That puts Astra at exactly the price Anthropic set for Claude Fable 5.2 a week earlier.
What is new in GPT-6.2 Astra and Sol
OpenAI says Astra was trained to follow professional conventions in law, finance, sales and accounting more closely, for example matching a slide template’s layout, colors and font sizes from a file or screenshot. It also claims better computer use. Sol is pitched as the default for everyday coding and agents, and OpenAI says paid users get more tasks done with it on the same plan than with GPT-6.1 Sol.
For developers, OpenAI’s migration guide flags two changes. The Responses API now drops earlier reasoning items when conversation history is edited mid-way or the model is switched, which can lower answer quality. Top-level instructions are appended after the input, so reusable instructions belong in a first developer message to stay cacheable.
Pricing and availability
Both models are rolling out to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex; on Enterprise and Edu, a workspace admin must turn them on. In the API they are gpt-6.2-astra and gpt-6.2-sol. Cached input costs $0.40 per million tokens on Astra and $0.10 on Sol. Per OpenAI’s model pages, requests above 272,000 input tokens pay double the input rate and 1.5 times the output rate, and Batch costs half.
OpenAI also said Astra for Law, a legal version of Astra, ranked first in a launch evaluation spanning 24 areas of law, and that every ChatGPT Enterprise customer can now request access.
OpenAI’s benchmark claims and early outside checks
These figures are OpenAI’s own. It says Astra scores 5.2 points above GPT-6 Astra on Agents’ Last Exam, a multi-step professional work test, at under a third of the cost, and 5.5 points above Fable 5.2 at 5 percent of its cost. It claims a 9-point gain on Terminal-Bench-Science 0.1 and more than double Fable 5.2’s pass rate on the Harvey LAB-AA legal test. Sol, OpenAI says, lands within 0.5 points of Astra on FrontierCode 2.0 at about a fifth of the cost.
Outside evaluators are more mixed. The Decoder reports that Artificial Analysis puts Astra second on its Intelligence Index with 62 points, behind Fable 5.2 at 65, with Sol at 57, but sees no coding gain. Vals AI ranks Astra 7th and Sol 9th while finding weak legal research results. Simon Willison called early reaction mixed, citing complaints about confusing questions from Astra on long tasks.
How GPT-6.2 compares with GPT-6 Astra and GPT-6.1 Sol
The new pair changes price and freshness more than shape. Both keep a 1.05 million-token context window, 922,000 maximum input tokens and 128,000 output tokens, the same as GPT-6 Astra and GPT-6.1 Sol. The knowledge cutoff moves from April 30, 2026 to May 18 for Astra and June 1 for Sol.
| Model | Input / output per 1M tokens | Cached input | Knowledge cutoff |
|---|---|---|---|
| GPT-6 Astra | $10 / $50 | $1.00 | Apr 30, 2026 |
| GPT-6.2 Astra | $8 / $40 | $0.40 | May 18, 2026 |
| GPT-6.1 Sol | $2 / $10 | $0.10 | Apr 30, 2026 |
| GPT-6.2 Sol | $2 / $10 | $0.10 | Jun 1, 2026 |
The 60 percent cut in cached-input price matters most for long agent sessions that reread the same context. With Astra now matching Claude Fable 5.2’s $8/$40 pricing, token use and task results decide between them.
Safety ratings
Like GPT-6 Astra, both models reach the Critical cybersecurity level under OpenAI’s Preparedness Framework, with stricter safeguards that OpenAI says also limit some legitimate security research. SiliconANGLE reports that Astra is the first externally released OpenAI model at the High level for AI self-improvement. In two alignment tests the models more often failed to mention obstacles such as a missing file.
What to watch next
- When the GPT-6.2 models reach OpenAI’s Daybreak program for vetted cyber defenders.
- Independent coding results, after Artificial Analysis found no coding gain.
- OpenAI’s follow-up on models omitting obstacles, which it says it is investigating.
- Whether Anthropic or Google answers the price match with new cuts.
Key takeaways
- GPT-6.2 Astra and Sol launched on November 17 in ChatGPT Work, Codex and the API.
- Astra costs $8/$40 per million tokens, 20 percent below GPT-6 Astra, with cached input down to $0.40.
- Sol keeps the $2/$10 price of GPT-6.1 Sol.
- OpenAI’s benchmark gains are self-reported; outside evaluators rank Astra just behind Claude Fable 5.2.
- Both models carry a Critical cybersecurity rating and tighter safeguards.
Frequently asked questions
How much does GPT-6.2 Astra cost?
In the API it costs $8 per million input tokens, $0.40 for cached input and $40 for output. Requests above 272,000 input tokens cost more.
Who can use GPT-6.2 Astra and Sol in ChatGPT?
Plus, Pro, Enterprise and Edu users can use both models in ChatGPT Work and Codex. Enterprise and Edu workspaces need an admin to enable them.
Is GPT-6.2 Sol better than GPT-6.1 Sol?
OpenAI says it does more work per plan at the same price and comes close to Astra on coding. Artificial Analysis, however, measured a small coding drop, so results depend on the task.




