OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November 2026, ten weeks after GPT-6 Astra arrived in September. Astra is the model for the hardest professional work; Sol is the cheaper everyday model for coding, writing and multi-step agent tasks. The headline changes are a 20% cut to Astra’s API price and claims that both models finish demanding jobs with fewer output tokens. The system card also contains the most serious safety ratings OpenAI has published for a released model.
What OpenAI released
According to OpenAI’s developer announcement, both models are available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.2-astra and gpt-6.2-sol. Rollout is staged by plan and workspace, so some users will not see the models immediately.
OpenAI also expanded its Ultrafast mode, which generates tokens up to 4.5 times faster than the standard rate but uses more of a plan’s allowance. CASRAI’s summary reports that the $500-a-month Pro tier gets Ultrafast on both models, the $200 Pro tier starts with Sol, and Enterprise and Edu admins can switch it on.
GPT-6.2 pricing
| Model (per 1M tokens) | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Astra | $8.00 | $0.40 | $40.00 |
| GPT-6.2 Sol | $2.00 | $0.10 | $10.00 |
| GPT-6 Astra (previous) | $10.00 | $1.00 | $50.00 |
Astra is 20% cheaper than GPT-6 Astra on input and output and 60% cheaper on cached input, while Sol’s price is unchanged. OpenAI says Astra costs less than a third of GPT-6 Astra per completed task because it uses fewer tokens. That per-task figure is OpenAI’s own claim and will vary with workload, so teams should test it against their own jobs before switching.
What improved
- Agentic work: OpenAI reports that Astra at its xhigh setting scores 5.2 points above GPT-6 Astra at max on Agents’ Last Exam, at less than half the cost, as 9to5Mac reported.
- Coding in Codex: early testers described cleaner code style, more focused tests and fewer revision rounds. On FrontierCode 2.0, which checks whether changes to real codebases would be merged, OpenAI says Sol lands within half a point of Astra at about one-fifth of the cost.
- Science: Vals AI ranks Astra first on its Science Index, and Epoch AI reports Astra solved 7 of 68 FrontierMath Erdős problems, each verified in Lean, against a previous record of 3.
What independent testers found
Independent results are more mixed than the launch materials. Artificial Analysis places Astra second on its Intelligence Index with a score of 62, up five points on GPT-6 Astra and behind Anthropic’s Claude Fable 5.2 at 65; Sol scores 57. The same group found Astra’s coding score flat and Sol’s three points lower than GPT-6.1 Sol. On its AA-Omniscience test of very hard factual questions, Astra hallucinated 59% of the time at maximum effort and Sol 55%.
Epoch AI found that on its real-world research tasks Astra trails Claude Fable 5.2 and barely improves on GPT-6 Astra. The Rundown described early reviews as rough despite the benchmark gains.
The safety findings
The system card is the most important document in this release. Both models meet OpenAI’s Critical threshold for cybersecurity capability and the High threshold, but not Critical, for biological and chemical capability. Astra is the first externally released OpenAI model to reach the High threshold for AI self-improvement; Sol does not.
In practice, OpenAI has tightened default restrictions on higher-risk cyber assistance. More permissive access requires extra vetting through its Trusted Access for Cyber programme, and the company says it is working to bring the models to Daybreak, where approved defenders get fewer refusals. Critics note that the strongest cyber results come from internal tests with safeguards switched off, and that stricter defaults may also slow legitimate security research.
Should you switch?
For API users already on GPT-6 Astra, the lower price makes GPT-6.2 Astra the obvious default to evaluate first. For high-volume coding and writing, Sol is the better value if its quality holds up in your own tests. Anyone relying on factual accuracy should still verify outputs, as the hallucination results show. Our guide to fact-checking AI answers applies to any model, and our Claude Fable 5.2 vs GPT-6.2 Astra comparison covers how the two flagships compare.
FAQ
When was GPT-6.2 released?
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November 2026.
How much does GPT-6.2 Astra cost?
$8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens.
Is GPT-6.2 available on the free ChatGPT plan?
No. At launch it is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through the API.
Sources
- OpenAI Developer Community: GPT-6.2 Astra and Sol are now available
- 9to5Mac: OpenAI upgrades ChatGPT and Codex with GPT-6.2
- CASRAI: GPT-6.2 pricing and system card summary
- The Rundown: early reviews of GPT-6.2
Published
Spotted an error? Read our corrections policy or email editorial@perplexityaimagazine.com.