# OpenAI's GPT-6.2 Sol matches its earlier flagship at about 15 percent of the cost per task

GPT-6.2 Astra lists at $8 per million input tokens and $40 per million output tokens, 20 percent below GPT-6 Astra. Independent indexes still put Anthropic's Claude Fable 5.2 ahead on broad agentic and terminal tasks.

- Published: 2026-11-22T06:11:42.557Z
- Canonical: https://polylog.news/ai/2026-11-22/openai-s-gpt-6-2-sol-matches-its-earlier-flagship-at-about-1
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [OpenAI Research](https://openai.com/index/introducing-gpt-6-2-astra-sol/), [OpenAI Research](https://openai.com/index/gpt-6-astra/), [OpenAI Research](https://openai.com/index/introducing-gpt-6-1-sol/), [Artificial Analysis](https://artificialanalysis.ai/articles/openai-has-the-2-model-in-the-intelligence-index-with-the-release-of-gpt-6-2-astra-and-gpt-6-2-sol), [4sysops](https://4sysops.com/archives/gpt-6-2-astra-and-sol-review-benchmarks-and-early-tester-bashing/), [DataCamp](https://www.datacamp.com/blog/gpt-6-2-astra-and-sol), [OpenAI Deployment Safety Hub](https://deploymentsafety.openai.com/gpt-6-2-astra-sol)

OpenAI [released GPT-6.2 Astra and GPT-6.2 Sol on November 17](https://openai.com/index/introducing-gpt-6-2-astra-sol/). The release mainly lowers price while leaving peak capability about the same. On the Artificial Analysis Intelligence Index, Sol [scores 57](https://artificialanalysis.ai/articles/openai-has-the-2-model-in-the-intelligence-index-with-the-release-of-gpt-6-2-astra-and-gpt-6-2-sol), matching [GPT-6 Astra](https://openai.com/index/gpt-6-astra/), the flagship OpenAI shipped on September 3, while costing about 15 percent as much per task. The release puts OpenAI's models second on the index.

Reported list prices are $8 input and $40 output per million tokens for Astra, and $2 and $10 for Sol, with a 1.05-million-token context window ([Braindetox](https://braindetox.kr/en/posts/gpt_6_2_astra_sol_release.html)). Sol's price is unchanged from [GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/), which launched on September 29.

On coding, OpenAI's own figures put Sol at 43.8 percent on FrontierCode 2.0 and Astra at 44.2 percent. The cost per task is $1.29 for Sol and $5.97 for Astra ([DataCamp](https://www.datacamp.com/blog/gpt-6-2-astra-and-sol)). Artificial Analysis's own measurements show smaller gains: Sol lost three points on its Coding Agent Index compared with GPT-6.1 Sol, and Astra did not improve on GPT-6 Astra. Vals AI's Vals Index puts Astra at 67.66 percent against 71.31 percent for Claude Fable 5.2. Fable also leads on Terminal-Bench 4.0, 65.15 percent to 63.13 percent. Astra leads on the Vals Science Index, 68.44 percent to 63.74 percent ([4sysops](https://4sysops.com/archives/gpt-6-2-astra-and-sol-review-benchmarks-and-early-tester-bashing/)).

Two numbers matter for anyone deploying these models. Artificial Analysis measured hallucination rates of 59 percent for Astra and 55 percent for Sol at maximum effort. OpenAI's [system card](https://deploymentsafety.openai.com/gpt-6-2-astra-sol) reports that Astra solved 36.4 percent of targets on a critical biology evaluation, up from 18.2 percent for GPT-6 Astra. That doubling in a dual-use domain is the clearest capability jump in the whole release.

Google priced Gemini 4 Neon at [$3 input and $15 output](https://www.datacamp.com/blog/gemini-4-neon) the next day. Every major lab now has a model near the top of the frontier priced well below its flagship from two months earlier.

## What this means

The cost of frontier-level output is falling faster than top scores are rising, which helps application builders and squeezes API margins at OpenAI, Anthropic and Google. The desk expects another round of price cuts within weeks, assuming GPU supply holds up. A lab raising prices would show the price war has ended.

## What to watch

- Whether Anthropic cuts Fable 5.2 prices further in response to Sol. A cut would confirm that labs now compete for share on price.
- Independent replication of the biology capability increase reported in OpenAI's system card. A doubling in a dual-use domain is the kind of result that leads to gated releases.
- Whether the reported 55 to 59 percent hallucination rates fall in the next point release. Those rates limit how much enterprises can automate legal and research work.
