Skip to content

GPT-6.2 Astra and Sol: price, benchmarks and access.

OpenAI’s GPT-6.2 Astra and GPT-6.2 Sol are out in ChatGPT Work, Codex and the API. Prices, context window, benchmarks and how to try them.

By The Model Atlas · 4 sources

In this post8 sections
  1. 01What are GPT-6.2 Astra and GPT-6.2 Sol?
  2. 02What’s new in GPT-6.2 Astra
  3. 03What’s new in GPT-6.2 Sol
  4. 04GPT-6.2 Astra and Sol pricing
  5. 05How to try GPT-6.2 Astra and Sol
  6. 06Astra or Sol: which should you use?
  7. 07What people are building with GPT-6.2
  8. 08GPT-6.2 FAQ

GPT-6.2 Astra and GPT-6.2 Sol are OpenAI’s newest models, launched on 17 November 2026. OpenAI calls GPT-6.2 Astra its new frontier model for your hardest projects and GPT-6.2 Sol its efficient, versatile model for everyday work. Both are available today in ChatGPT Work, Codex and the OpenAI API. Here is what GPT-6.2 Astra and Sol are, what OpenAI says is new, what they cost and how to try them.

What are GPT-6.2 Astra and GPT-6.2 Sol?

GPT-6.2 is a two-model release. Astra is the larger model, aimed at complex professional work: research, long coding jobs, slides, spreadsheets and documents. Sol is the faster, cheaper model for everyday coding, writing and multistep agent work. They follow GPT-6 Astra and GPT-6.1 Sol, the earlier versions OpenAI compares them against.

OpenAI’s API docs list the same core specs for both models:

  • Text and image input, text output (no audio or video).
  • A 1,050,000-token context window, up to 922,000 input tokens and 128,000 output tokens.
  • Reasoning token support. The API rejects requests that set reasoning effort to none.
  • Knowledge cutoff of 18 May 2026 for Astra and 1 June 2026 for Sol.
  • Model IDs gpt-6.2-astra and gpt-6.2-sol, on the Chat Completions, Responses and Batch endpoints.

What’s new in GPT-6.2 Astra

OpenAI says GPT-6.2 Astra produces slides, spreadsheets and documents that read less like AI and more closely follow your templates and the conventions of fields such as law, sales, finance and accounting. It also says Astra writes cleaner code and builds better-looking frontends, particularly for websites, games and 3D design, and that its more precise computer use lets it work down to individual parts in CAD and simulation tools.

The headline benchmark claims, all from OpenAI’s announcement:

  • Agents’ Last Exam (multistep professional work): a new high, 5.2 points above GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at 5% of the cost.
  • Terminal-Bench-Science 0.1: a new high, 9 percentage points above GPT-6 Astra. OpenAI says it believes Astra is the best model in the world for scientists.
  • GDPval v2 (OpenAI’s internal long-horizon knowledge-work test): up 6.2 points from GPT-6 Astra while using 22% fewer output tokens.
  • Harvey LAB-AA v1.1 (legal work): a new state of the art, more than doubling Claude Fable 5.2’s pass rate at about one-sixth of the cost.
  • AA-AnalystAgent from Artificial Analysis: 62.5% of questions answered correctly on all five attempts.
  • AA-3D from Artificial Analysis, where human judges vote between blinded 3D experiences: Astra first at 1,318 Elo, then Claude Fable 5.2 at 1,271 and GPT-6.2 Sol at 1,259.

OpenAI also says GPT-6.2 Astra is its first externally released model to meet the High threshold for AI self-improvement under its Preparedness Framework, and that both models meet the Critical threshold for cybersecurity. Because of that, it is launching them with tighter safeguards that, OpenAI notes, also restrict some legitimate security-testing work.

What’s new in GPT-6.2 Sol

OpenAI says GPT-6.2 Sol is smarter and more efficient than GPT-6.1 Sol and lets you finish more tasks on the same paid plan. It says many OpenAI employees use Sol as their default model for both technical and non-technical work.

  • On Cognition’s FrontierCode 2.0, OpenAI says Sol comes within 0.5 points of GPT-6.2 Astra at about one-fifth of the cost (both at max reasoning).
  • On OpenAI’s LifeVal everyday-task test, it says Sol outscores GPT-6 Astra at about one-seventh of the cost.
  • On GDPval v2, OpenAI reports Sol up 5.3 points from GPT-6.1 Sol.

GPT-6.2 Astra and Sol pricing

OpenAI’s standard API prices per million tokens:

  • GPT-6.2 Astra: $8 input, $0.40 cached input, $40 output. OpenAI says that is 20% less than GPT-6 Astra for input and output and 60% less for cached input.
  • GPT-6.2 Sol: $2 input, $0.10 cached input, $10 output, the same rates as GPT-6.1 Sol.

The API docs add a few details. Requests with more than 272K input tokens pay double input and cache rates and 1.5 times the output rate for the whole request. Batch costs 50% of Standard, Fast mode costs twice Standard, and Ultrafast costs three times Standard through 28 February 2027 and six times Standard from 1 March 2027. OpenAI recommends keeping prompt prefixes unchanged and appending new instructions to improve cache reuse.

How to try GPT-6.2 Astra and Sol

  1. ChatGPT Work: OpenAI says both models are available to Plus, Pro, Enterprise and Edu users, and ChatGPT Learn also lists Business. Enterprise and Edu admins must enable them first. On supported consumer plans, new Work chats default to Sol at Light reasoning; open Advanced to pick a model and effort. To use them from ChatGPT, open Work.
  2. Codex: choose Astra or Sol in the Codex app’s model picker or with /model in the CLI. ChatGPT Learn suggests Astra for tracing a bug across services and Sol for reviewing a pull request. Codex and Work share usage.
  3. API: call gpt-6.2-astra or gpt-6.2-sol in the Responses API or Chat Completions. Both support Ultrafast at launch.
  4. Ultrafast in ChatGPT: Pro $500 subscribers can use it with either model, and Pro $200 subscribers with Sol.

Astra or Sol: which should you use?

OpenAI positions Astra for the hardest, most complex reasoning, such as working through a research question across several sources, building a deck from a report in your template, or fixing a bug that spans services. Sol is pitched for everyday coding, writing and multistep work, such as reviewing a pull request, revising code after feedback or turning meeting notes into a brief. Given OpenAI’s FrontierCode 2.0 result and the 4x price gap, Sol is a sensible default, with Astra kept for jobs where quality matters most.

What people are building with GPT-6.2

Builders have already started testing the models. Two early examples on The Model Atlas:

See the GPT-6.2 Astra and GPT-6.2 Sol model pages for new builds as they are added. For the rival model OpenAI compares against most, read our Claude Fable 5.2 explainer.

GPT-6.2 FAQ

When was GPT-6.2 released?

OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on 17 November 2026.

How much does GPT-6.2 Astra cost?

$8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens at standard API rates, according to OpenAI.

How much does GPT-6.2 Sol cost?

$2 per million input tokens, $0.10 cached input and $10 output, the same as GPT-6.1 Sol.

What is the GPT-6.2 context window?

OpenAI’s API docs list a 1,050,000-token context window for both models, with up to 922,000 input tokens and 128,000 output tokens.

Is GPT-6.2 available in free ChatGPT?

OpenAI lists availability for paid plans: Plus, Pro, Enterprise and Edu (ChatGPT Learn also lists Business), in ChatGPT Work and Codex. It doesn’t list the free plan.

What are the GPT-6.2 API model names?

gpt-6.2-astra and gpt-6.2-sol.

Can GPT-6.2 generate images or audio?

Not directly. Both models take text and image input and return text, though the Responses API lists image generation among the tools they can call.

Is GPT-6.2 Astra better than Claude Fable 5.2?

On OpenAI’s reported results it leads Fable 5.2 on Agents’ Last Exam, Harvey LAB-AA and AA-3D, at a fraction of the cost. These are OpenAI’s own comparisons; check independent tests for your use case.

Sources · 4

  1. Introducing GPT-6.2 Astra and Sol (OpenAI, 17 November 2026) (opens in a new tab)
  2. GPT-6.2 Astra model page (OpenAI API docs) (opens in a new tab)
  3. GPT-6.2 Sol model page (OpenAI API docs) (opens in a new tab)
  4. GPT-6.2 (ChatGPT Learn, 17 November 2026) (opens in a new tab)
All 8 builds →