OpenAI has released GPT-6.2 Astra, for harder professional work, and GPT-6.2 Sol, for everyday tasks. OpenAI says both get better results with fewer output tokens. The API names are gpt-6.2-astra and gpt-6.2-sol.
The system card is dated 17 November 2026 and says, “Today, we are releasing.” Checked 18 November 2026 (Beijing time).
Who can use them?
The announcement says Plus, Pro, Enterprise, and Edu users can use both models in ChatGPT Work and Codex starting today. Pro 500 subscribers can use Ultrafast with either model. Ultrafast is reaching Pro 200 subscribers for the first time, starting with GPT-6.2 Sol. Eligible Enterprise and Edu users get the same access once a workspace admin enables it. Developers can call both models in the API, with Ultrafast available at launch. The announcement does not say whether the Free plan is included. This article relies only on official OpenAI channels and does not mean Orange Relay or another third-party route already offers these models.
What do they cost?
Standard API prices are per 1 million tokens and are not the whole bill.
GPT-6.2 Astra: $8 input, $0.40 cached input, and $40 output. The announcement says that is 20% lower than GPT-6 Astra on input and output, with a 60% discount on cached input.
GPT-6.2 Sol: the announcement says it keeps GPT-6.1 Sol’s API prices and restates them—$2 input, $0.10 cached input, and $10 output. The official pricing and model pages list the same standard short-context rates, matching GPT-6.1 Sol.
Other tiers are on the pricing page. Above 272K input tokens, input and cache rates double and output costs 1.5× for the full request; Batch is 50% of Standard; Fast is 2×; Ultrafast is 3× through 28 February 2027, then 6× from 1 March 2027 Pacific time. Cache writes are extra.
What does OpenAI claim?
OpenAI positions Astra as its frontier model for the hardest projects, and Sol as the everyday default. On Agents’ Last Exam, it says GPT-6.2 Astra scores 5.2 points higher than GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at about 5% of the cost; Fable 5.2’s leaderboard results exclude nine tasks that used a fallback model. These are OpenAI’s own claims; we have not independently rerun them.
What do the safety ratings mean?
Under OpenAI’s Preparedness Framework, both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities. OpenAI calls GPT-6.2 Astra its first externally released model to meet the High threshold for AI self-improvement; GPT-6.2 Sol falls below that threshold. It explains High as an impact equivalent to giving every OpenAI researcher an extra, highly capable research-engineer assistant, relative to a 2024 baseline. That is not a claim that the model is out of control. The system card rates both models at Loss of Control Tier 1 under OpenAI’s Frontier Governance Framework, because they do not yet reliably evade detection across varied evaluations.
Our take
Developers should first check official prices and whether their plan can use the models in ChatGPT Work, Codex, or the API. Astra’s official cut versus GPT-6 Astra is the published standard input/output price and a deeper cached-input discount. Sol’s standard price, OpenAI says, stays at GPT-6.1 Sol’s rate and has been restated. Critical and High are capability thresholds, not a claim that the models have gone out of control. We have not production-tested these models.
Sources: