# OpenAI prices GPT-6.2 Astra at $8 and $40 per million tokens, the same list rate as Anthropic's Claude Fable 5.2

GPT-6.2 Sol keeps GPT-6.1 Sol's $2 and $10 rates and scores within half a point of Astra on the FrontierCode 2.0 coding benchmark, according to third-party compilations of OpenAI's figures.

- Published: 2026-11-21T06:18:37.672Z
- Canonical: https://polylog.news/ai/2026-11-21/openai-prices-gpt-6-2-astra-at-8-and-40-per-million-tokens-t
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [OpenAI Research](https://openai.com/index/introducing-gpt-6-2-astra-sol/), [OpenAI News (Hebbia)](https://openai.com/business/customer-stories/hebbia/), [Requesty](https://www.requesty.ai/blog/gpt-6-2-sol-astra-pricing-benchmarks)

OpenAI released [GPT-6.2 Astra and GPT-6.2 Sol](https://openai.com/index/introducing-gpt-6-2-astra-sol/) on November 17. Astra lists at $8 per million input tokens, $0.40 for cached input and $40 for output. Several pricing trackers, including [Requesty](https://www.requesty.ai/blog/gpt-6-2-sol-astra-pricing-benchmarks), report that this is 20% below GPT-6 Astra. Sol keeps GPT-6.1 Sol's $2 input, $0.10 cached input and $10 output rates. Prompts over 272,000 tokens cost more, with Astra rising to $16 input and $60 output, [according to one pricing table](https://computingforgeeks.com/gpt-6-2-astra-sol-pricing-api/).

Anthropic launched Claude Fable 5.2 on November 9 at [$8 input and $40 output](https://www.eesel.ai/blog/claude-mythos-5-2). Secondary coverage reports that this was [20% below Fable 5.1](https://techenclave.com/news/items/anthropic-releases-claude-fable-5-2-at-lower-api-prices). Within eight days, both labs cut their flagship list price and arrived at the same headline rate.

Most of the capability claims come from OpenAI. On its internal GDPval v2 evaluation, the company reports that Astra gains about 6.2 points over GPT-6 Astra while using 22% fewer output tokens. On Terminal-Bench-Science 0.1, OpenAI reports a gain of about nine points at under a third of the cost. Independent numbers are thinner. One aggregator reports [Artificial Analysis Intelligence Index scores](https://www.requesty.ai/blog/gpt-6-2-sol-astra-pricing-benchmarks) at maximum effort of 65 for Fable 5.2, 62 for Astra and 57 for Sol, putting Anthropic's model ahead on that composite. Artificial Analysis also flagged a regression for Sol on a repository question-answering test.

The more useful number for engineers is the gap between OpenAI's two tiers. On FrontierCode 2.0, Sol scores 43.8% and Astra 44.2%, with mean listed costs per run of $1.29 and $5.97, [according to DataCamp's compilation](https://www.datacamp.com/blog/gpt-6-2-astra-and-sol). OpenAI also published a [Hebbia customer story](https://openai.com/business/customer-stories/hebbia/) saying Hebbia's prerelease evaluations of GPT-6 Astra for financial research found gains in source coverage, citation relevance and presentation quality. The customer story gives no figures, and the vendor published it.

## What this means

Two vendors cut flagship list prices by 20% within eight days and converged on the same $8 and $40 rate, each setting its price to match the other's. Cheaper tiers such as Sol will likely take most coding traffic within months, unless independent tests confirm Sol's repository-task regression on real codebases.

## What to watch

- Google's next Gemini price change. A cut below $8 and $40 would confirm that the three largest closed labs are setting prices against each other.
- Independent coding results for Sol on large private repositories. These would show whether the Artificial Analysis regression matters in production.
- Usage mix reported by API routers between Astra and Sol. A shift toward Sol would show that buyers are trading the top tier for cost.
