OpenAI shipped GPT-6.2 Astra and GPT-6.2 Sol two days before Anthropic's Nasdaq debut, matching Anthropic's flagship pricing rather than undercutting it. Astra is priced at $8 per million input tokens and $40 per million output, while Sol remains at $2 input and $10 output.
OpenAI shipped GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, two days before Anthropic's Nasdaq debut, and the pricing table tells the sharper story than the benchmarks do.
Astra, the frontier model for long professional work, costs $8 per million input tokens and $40 per million output in the API, with cached input at $0.40. That is exactly what Anthropic lists for Claude Fable 5.2 and Mythos 5.2, which launched eight days earlier. Sol, the everyday model, carries forward GPT-6.1 Sol's rates unchanged at $2 input and $10 output, the same band as Claude Sonnet 5.5.
The undercutting happened one generation back. Fortune reported that Anthropic's November 9 releases moved below the price of the previous GPT-6 Astra. OpenAI's answer was not to go under Anthropic's flagship tier but to match it while cutting 20% from its own previous generation and taking 60% off cached input.
What OpenAI claims
The efficiency argument rests on tokens per task rather than token prices. OpenAI says Astra costs less than a third of what GPT-6 Astra cost per task, and that it beat Claude Fable 5.2 by 5.5 points on Agents' Last Exam at 5% of the cost. Both models also set company highs on Terminal-Bench-Science and the Vals Science Index, part of OpenAI's pitch to be "the best model in the world for scientists."
All of those are maker-reported numbers. On the independent Artificial Analysis index, Claude Fable 5.2 at max effort still leads Astra on several measures. Treat the cost-per-task gap as OpenAI's claim until third parties size it.
The access caveats

Launch availability is narrower than a typical frontier release. Both models run in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and through the API as gpt-6.2-astra and gpt-6.2-sol. OpenAI's help center confirms the rollout does not touch regular Chat, and Free and Go plans get nothing. Ultrafast tier access starts at Pro 500 for both models and, for the first time, Pro 200 for Sol.
Reception among heavy users was poor enough for VentureBeat to run "Essentially unusable" as the lede, citing complaints about limits and behavior in the days after launch.
The second attempt
This is OpenAI's second run at the 6.2-class model. The company scrapped GPT-6.1 Astra in late September after internal tests flagged failures around staying within its authorized scope and accurately reporting the work it had done. OpenAI's own launch notes acknowledge related safety notes, which Dealroom's coverage called as notable as the scores.
The calendar is the final fact worth stating. A frontier model priced to parity, released during a rival's roadshow, with benchmark claims the rival's own models still contest on independent leaderboards. Competition in this market now expresses itself in release timing as much as in capability.
Sources:
- OpenAI: Introducing GPT-6.2 Astra and Sol
- VentureBeat: OpenAI launches GPT-6.2 Astra and Sol to negative reception
- MLQ.ai: OpenAI launches GPT-6.2 Astra and Sol, cutting Astra's API price by 20%
- Dealroom: OpenAI undercuts Claude with GPT-6.2 Astra in Anthropic's IPO week
Frequently asked questions
How does GPT-6.2 Astra's API pricing compare to Anthropic's models?
GPT-6.2 Astra is priced at $8 per million input tokens and $40 per million output, which is exactly the same as Anthropic's Claude Fable 5.2 and Mythos 5.2.
Who has access to GPT-6.2 Astra and Sol?
Both models are available in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users. They are not accessible on regular Chat, Free, or Go plans.
What are the pricing tiers for Ultrafast access to these models?
Ultrafast tier access starts at Pro 500 for both GPT-6.2 Astra and Sol. For GPT-6.2 Sol specifically, Ultrafast access also begins at the lower Pro 200 tier.
Why was GPT-6.1 Astra scrapped before this launch?
OpenAI scrapped GPT-6.1 Astra in late September after internal tests flagged failures regarding staying within authorized scope and accurately reporting work performed.
