Bookmarks

    Nothing saved yet. Use the bookmark on any article. Bookmarks are kept on this device only: they stay private and work without an account.

    Settings

    Appearance

    Reading mode

    Font size

    Content width

    Line spacing

    Keyboard shortcuts

    /
    Search
    j k
    Next / previous story
    ← →
    Previous / next section (columns)
    b
    Bookmarks
    s
    Settings
    r
    Columns / scroll
    t
    Light / dark

    Saved in this browser only.

    Ping2Prompt
    Subscribe to the Weekly Brief Sign in
    Translucent blue glass rods rise behind the gray 3D letters spelling OpenAI on a plain gray background.
    Photo by Brecht Corbeel on Unsplash

    OpenAI releases GPT-6.2 Astra and Sol, with Astra API prices cut

    Automated source check

    Hugging Face OpenAI

    Update (2026-11-18): This article now covers everything published about the GPT-6.2 release (18 sources), merged from separate reports.

    OpenAI ships GPT-6.2 Astra and Sol on 17 November

    OpenAI released GPT-6.2 Astra and Sol on 2026-11-17. GitHub Copilot made both models generally available the same day.

    Astra was the first GPT-6 model, out in September, and Sol moved to GPT-6.1 at DevDay a few weeks after that.

    What Astra and Sol are each meant for

    OpenAI positions Astra as its frontier model for the hardest projects and Sol as the efficient, versatile model for everyday work. Decrypt lists spreadsheets, legal research and interactive 3D design as Astra's target jobs.

    OpenAI claims gains for both models in professional-grade output, coding, computer use and efficiency, at reduced cost, across ChatGPT Work, Codex and the API. It also says both finish demanding tasks with fewer output tokens, so you hit usage limits later.

    OpenAI says Astra sticks more closely to slide templates and spreadsheet conventions, with less filler and steadier layouts. That is the company's claim, not a tested result. Early GPT-6 Astra users complained in September that writing and creative quality had dropped, and OpenAI says this release addresses that.

    Scores, tester reports and OpenAI's risk ratings for GPT-6.2

    Artificial Analysis, an independent evaluator, gives Astra 62 on its Intelligence Index, a 100-point score that combines nine tests. It scores Anthropic's Claude Fable 5.2 at 65, which puts Astra second. Sol scores 57 there, matching the original GPT-6 Astra at roughly 15% of its cost per evaluated task.

    Independent tests that Decrypt reports show no gain in Astra's coding-agent performance. GitHub's early testing found Astra caught significantly more code review issues than GPT-6 Astra, using fewer tokens on agentic coding tasks. GitHub also reports that Sol found nearly twice as many real review issues as GPT-6.1 Sol, with findings still accurate.

    Testing found the models sometimes failed to say when missing files or unavailable tools blocked their work. Early testers also reported unproductive loops, tasks declared done too soon and confusing replies. OpenAI believes some reviewers missed last-minute improvements, and it concedes the user experience can still get better.

    OpenAI says Astra is the first publicly released model rated High for AI self-improvement, and the company itself counts that as a risk. Decrypt reports that, per OpenAI, both models meet the company's Critical cybersecurity threshold. OpenAI says it tightened monitoring and added a system named Control Room that can stop suspicious activity by the models. GPT-6.2 is not yet part of OpenAI's Daybreak program for vetted cybersecurity researchers.

    API IDs, token prices and where GPT-6.2 runs

    In the API, call the models as gpt-6.2-astra and gpt-6.2-sol. Pricing differs by model:

    • Astra's cached input rate is $0.40 per million tokens, against $1 for the older model.
    • Sol stays at $2 per million input tokens and $10 per million output tokens.

    Per Decrypt, OpenAI offers the models to Plus, Pro, Enterprise and Edu plans in ChatGPT Work and Codex, and through the API. A GPT Plus subscription page did not yet list them when Decrypt reported.

    In GitHub Copilot, the models are open to Pro+, Max, Business and Enterprise users, with a gradual rollout. They are billed at provider list pricing under usage-based billing. Business and Enterprise administrators can control access through the model policy in Copilot settings, and new models are on by default unless an administrator turns them off.

    Who should test GPT-6.2 and what to check

    If you pay for Astra by the token, rerun a sample of your own workloads and compare cost per task before switching. Test Sol against your current setup for everyday jobs, and have agent runs log when a tool or file is missing, since the models may not report it. Copilot admins should decide on the model policy setting before users start picking the new models.

    What OpenAI's announcement does not cover

    The OpenAI announcement text in the sources contains no prices or benchmark figures, and no source gives Sol's price from an OpenAI primary document.

    • No source gives an independent coding-agent score for the new models beyond the general index.
    • No source says when ChatGPT Plus subscribers will see the models.
    • The sources name Control Room but give no detail on how it works.

    Sources

    Image: Photo by Brecht Corbeel on Unsplash


    This story was produced with AI assistance from the sources listed above and fact-checked against them before publication. Read our editorial methodology and the corrections log. Spotted an error? Email [email protected].

    Continue

    Share on Bluesky · Mastodon · Email ·

    Discussion