OpenAI releases GPT-6.2 Astra and Sol: cheaper Astra, same-price Sol
What is GPT-6.2? On November 17, 2026 OpenAI released two new models: GPT-6.2 Astra for demanding reasoning and coding work, and GPT-6.2 Sol for everyday coding and agent workflows. According to The Decoder, Astra costs 8 dollars per million input tokens and 40 dollars per million output tokens in the API, 20 percent less than its predecessor. Sol costs 2 and 10 dollars, the same as GPT-6.1 Sol. Independent benchmarks put Astra in second place, behind Claude Fable 5.2.
We read OpenAI's API changelog and the report by The Decoder, which collects the prices and the results from independent evaluators. The launch spread on X through OpenAI's own account. The topic was also covered in the trending briefing of November 18.
What happened
The November 17 changelog lists the API models gpt-6.2-astra and gpt-6.2-sol. OpenAI recommends Astra for demanding reasoning and coding tasks and Sol for everyday coding and agent workflows. Both support the Responses API and Chat Completions, but tool calling requires the Responses API. GPT-6.2 also changes how the Responses API preserves reasoning when the start of a conversation is edited, and how the top-level instructions field is handled.
In ChatGPT, The Decoder reports, the models are available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. Pro-500 subscribers get the faster Ultrafast mode for both models; Pro-200 subscribers get it for Sol first.
A second changelog entry from the same day may matter just as much to developers. A public beta of the Responses API lets you pick a two-hour or eight-hour background completion window. Token rates are 30 percent (two hours) or 40 percent (eight hours) below the Standard rate. This applies to GPT-6.2 Astra and Sol, GPT-6.1 Terra and GPT-6 Luna.
Prices and benchmarks
| Item | GPT-6.2 Astra | GPT-6.2 Sol |
|---|---|---|
| Input per million tokens | 8 dollars | 2 dollars |
| Cached input | 0.40 dollars | 0.10 dollars |
| Output per million tokens | 40 dollars | 10 dollars |
| Artificial Analysis Intelligence Index | 62 (second, behind Fable 5.2 at 65) | 57 |
| Hallucination rate on a very hard question set | 59 percent | 55 percent |
| Cost per task according to Vals AI | 10.68 dollars | 3.18 dollars |
OpenAI says Astra beats Claude Fable 5.2 by 5.5 points on Agents' Last Exam at roughly 5 percent of the cost. Those are the maker's own figures. Artificial Analysis sees no coding gain for Astra and a 3-point drop for Sol. Epoch AI ties Astra with Fable 5.2 on its capabilities index but finds it behind on real-world research tasks. On FrontierMath Erdős, Astra solved 7 of 68 problems, up from a previous record of 3.
OpenAI says both models reach the Critical threshold for cybersecurity in its Preparedness Framework. It is therefore adding stricter safeguards for risky security requests, with vetted researchers getting access through its Daybreak programme. According to The Decoder, Astra is the first externally released model to reach OpenAI's High threshold for AI self-improvement.
What this means
The lower Astra price is a fact from the price list. The performance gain is mostly OpenAI's own claim, which independent evaluators only partly confirm. For heavy agent workloads Sol is the more interesting option: same price as GPT-6.1 Sol and, according to Artificial Analysis, the level of the old GPT-6 Astra at about 15 percent of the cost per task. Check it against your own usage data before switching, because Sol slipped slightly on coding.
The background windows offer a cheap route for work that does not need an instant answer, such as overnight evaluations or batch jobs. For more on saving tokens, see our article on context caching. On the previous generation, read our article on GPT-6.1 Terra; on thinking models, our overview of reasoning models; and on the security side, our overview of AI and cybersecurity.
What we did not verify
We did not read OpenAI's announcement page or its official pricing page ourselves; the prices come from The Decoder. The Decoder gives the 20 percent figure without the old price. A temporary Ultrafast discount appears only in a post by OpenAI's developer account on X. We did not rerun the benchmarks from OpenAI, Artificial Analysis, Epoch AI or Vals AI. Criticism on X, for example from @kimmonismus, is opinion.
Checked on November 18, 2026.
This article was written with AI (Claude) and checked against the sources below.
Sources
- OpenAI, API changelog (entries of November 17, 2026): developers.openai.com
- The Decoder, OpenAI launches GPT-6.2 Astra and Sol (November 17, 2026): the-decoder.com
- OpenAI on X about availability: x.com
- OpenAI Developers on X about API names and Ultrafast: x.com


