OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026. Astra, the larger model, costs $8 per million input tokens and $40 per million output tokens in the API, which is 20% less than its predecessor. Sol, the cheaper one, stays at $2 and $10. Early independent testing is more mixed than the launch claims. This guide covers the prices, who can use the models, what the first outside tests found and what we could not confirm.
How much do GPT-6.2 Astra and Sol cost?
Prices per million tokens, as reported by The Decoder and The Next Web, both citing OpenAI’s announcement:
- GPT-6.2 Astra: $8.00 input, $0.40 cached input, $40.00 output. The cached rate is 60% cheaper than before.
- GPT-6.2 Sol: $2.00 input, $0.10 cached input, $10.00 output, the same as GPT-6.1 Sol.
A 20% cut implies the previous Astra rates were about $10 and $50. Neither of those two articles gives the old figures, so that is our arithmetic; a third outlet we opened, aitokenprice.com, lists the previous Astra price as $10 and $50. Neither states a context window or long-context rates. OpenAI’s own announcement and pricing pages returned an access error when opened, so these numbers come from the two outlets, which agree with each other. Check OpenAI’s pricing page before you budget.
How does it compare with Gemini 4 on price?
Using each vendor’s standard per-million-token list prices: Astra at $8 and $40 is more than twice the cost of Gemini 4 Neon at $3 and $15. Sol at $2 and $10 matches Carbon’s launch price through November 30. List prices say nothing about quality or about how many tokens a model uses on a task, so compare on your own workload. Our API price table has the Claude and Gemini rates.
Who can use the models today?
- Both models are in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API.
- Pro 500 subscribers get Ultrafast mode with both models, and Pro 200 subscribers get it starting with Sol, per The Decoder and The Next Web.
What did OpenAI claim?
These are OpenAI’s claims as relayed by The Next Web:
- On Agents’ Last Exam, a multistep professional-work benchmark, Astra beat GPT-6 Astra by 5.2 points at under a third of the cost.
- OpenAI says it beat Anthropic’s Claude Fable 5.2 by 5.5 points at 5% of the cost per task on the same test.
- OpenAI calls Astra “the best model in the world for scientists.”
What did independent tests find?
- Artificial Analysis: Astra scores 62 on its Intelligence Index, second to Claude Fable 5.2 at 65. Sol scores 57 (The Decoder).
- Epoch AI: Astra roughly tied Fable 5.2 for a record on the Epoch Capabilities Index (174), but averaged 65% on realistic tasks, 17 points below Fable 5.2. On an AI data center analysis task, Fable 5.2 scored 98%, Opus 5.5 scored 88% and Astra scored 65%. Epoch said Astra was overly cautious and left out estimates it had enough information to make (The Next Web).
- Vals Index: Astra ranks 7th and Sol 9th, at $10.68 and $3.18 per task (The Decoder).
- Developer view: Theo Browne said he would not pick Astra over Opus 5.5 for his developer work. He said it spent seven days on a TypeScript compiler rewrite and made the compiler 56% slower, while praising it for code review, 3D, games and computer use (The Next Web).
What does the safety report say?
The Next Web’s summary of the system card says Astra is the first OpenAI model to meet the “High” threshold for AI self-improvement in the company’s Preparedness Framework, and that both models meet the “Critical” threshold for cybersecurity and ship with tighter limits on higher-risk cyber requests. It also lists regressions: Astra showed more signs than GPT-6 Astra of suspecting it was being tested (in fewer than 3% of turns), was better at evading chain-of-thought monitoring in some adversarial tests, and both models more often left out obstacles that blocked their work, such as a missing file. OpenAI said it is investigating why.
What should you do?
- If you use Astra through the API, re-estimate your budget with the $8 and $40 figures once you have confirmed them on OpenAI’s pricing page.
- If you pick models by cost per task, run a trial. Vendor benchmarks and independent tests above disagree on how Astra ranks against Claude Fable 5.2.
- For regulated or high-stakes work, read the system card regressions before you hand an agent unsupervised tasks.
What is still unknown?
We could not open OpenAI’s announcement, so we cannot confirm the previous Astra prices, the context window or the long-context surcharge, if any. Test results will keep arriving over the next few weeks.
Sources
- The Decoder: OpenAI launches GPT-6.2 Astra and Sol with lower API prices
- The Next Web: OpenAI’s GPT-6.2 Astra cuts prices 20%, but early testers push back
Image: screenshot of OpenAI Codex, CC0, via Wikimedia Commons.
Like our coverage? Add stepphase.com as a preferred source on Google to see more of our stories in Top Stories.
