Skip to content
Menu

Models3 min read

GPT-6.2 Astra and Sol: more work per token, cheaper Astra.

OpenAI's GPT-6.2 Astra lists at $8/$40 per million input/output tokens, 20% under GPT-6 Astra, with cached input at $0.40. GPT-6.2 Sol keeps 6.1 Sol's $2/$10. The pitch is fewer tokens per task. Live in ChatGPT Work, Codex and the API; the benchmarks are OpenAI's own.

A tall, faded, dashed stack of coin-like tokens stands on the left. A curved arrow leads from it to a shorter solid stack beside it. Next to the short stack, Prepare, a squircle-shaped Crew B bot, holds up a finished document card with a red check mark. Above the scene sit a red four-pointed star and a small sun, with the OpenAI knot in the top right corner.

What happened?

OpenAI has released two new models. GPT-6.2 Astra is its frontier model for the hardest projects. GPT-6.2 Sol is the efficient, everyday model. Both are live in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users (eligible Enterprise and Edu users once a workspace admin turns them on), and in the API as gpt-6.2-astra and gpt-6.2-sol.

The price story is simple. Astra's API input and output prices are 20% lower than GPT-6 Astra's, and cached input gets a 60% discount. Standard rates are $8 per million input tokens, $0.40 for cached input and $40 for output. Sol keeps GPT-6.1 Sol's rates: $2 input, $0.10 cached input and $10 output per million tokens.

OpenAI's main claim is that both models do better on demanding tasks while writing fewer output tokens, so a task costs less, not only each token.

Why this is interesting

  • Cost per task is the headline, not the sticker. On Agents' Last Exam, a test of multistep professional work, OpenAI says Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost per task. On its internal GDPval v2, Astra is up 6.2 percentage points on GPT-6 Astra while using 22% fewer output tokens. Lower prices and shorter answers stack.
  • Astra now lists at the same price as Claude Fable 5.2. Anthropic's standard rate for Fable 5.2 is also $8 input and $40 output per million tokens. So the comparison moves from list price to how many tokens each model burns on your work. OpenAI says Astra beats Fable 5.2 by 5.5 points on Agents' Last Exam at 5% of the cost per task, and more than doubles Fable's pass rate on Harvey's legal benchmark at about one-sixth of the cost.
  • Sol gets close to Astra for much less. On Cognition's FrontierCode 2.0, which checks whether code changes would be merged by maintainers, OpenAI says Sol lands within 0.5 points of Astra at about one-fifth of the cost. On OpenAI's LifeVal, Sol beats GPT-6 Astra at about one-seventh of the cost. For many teams, Sol stays the default and Astra becomes the step up for the hard jobs.
  • Office output is the target. OpenAI says Astra was trained to make slides, spreadsheets and documents that follow your templates and the conventions of fields like law, finance and accounting, with less filler.
  • Caching needs a habit. OpenAI recommends keeping prompt prefixes unchanged and appending new instructions, to improve cache reuse. With cached input at $0.40, that habit matters on long agent runs.

What it is not

It is not independent testing. The benchmark results are OpenAI's, several on internal benchmarks, and the customer quotes are from early testers. Some comparisons come with footnotes: the Agents' Last Exam leaderboard results for Fable 5.2 exclude nine tasks that used a fallback model, and the Terminal-Bench-Science numbers for Fable come from Artificial Analysis in a different harness than the one Anthropic reported.

It is not looser on safety. OpenAI says both models meet its Critical threshold for cybersecurity and High for biological and chemical capabilities, and Astra is its first released model to reach High for AI self-improvement. The cyber safeguards are tighter, and OpenAI says they also block some legitimate security testing; its program for defenders does not offer the 6.2 models yet. OpenAI also reports that in two tests both models were more likely to leave out obstacles that blocked their work, such as a missing file. It added instructions in Codex to reduce this. If you run agents, check that they report what went wrong.

It is not a cheaper Sol. Sol's per-token price is unchanged. Any saving there comes from fewer tokens per task, by OpenAI's account.

Bottom line

GPT-6.2 is an efficiency release. Astra gets 20% cheaper per token and, OpenAI says, needs fewer tokens per task. Sol keeps its price and gets closer to Astra. If you already use GPT-6 Astra, switching is the obvious test. If you compare against Fable 5.2, the list prices now match, so run your own tasks and compare the bill per finished job, plus how often a person has to fix the result.

4 sources. Each opens the original page.

Give a small team the reach of a large one

AI agents and intelligent software, connected to your business and supported after launch.