OpenAI Launches GPT-6.2 Astra and Sol (2026)

Abstract illustration of two model orbs beside document and code panels.

Written by

in

Takeaway: On November 17, 2026, OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol. Astra is the new frontier model for demanding professional work; Sol is the more efficient model for everyday tasks. Both are available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers through the API as gpt-6.2-astra and gpt-6.2-sol. OpenAI says Astra costs less per task than GPT-6 Astra, with API rates 20% lower on input and output and a 60% discount on cached input: $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens. Sol keeps GPT-6.1 Sol’s rates of $2 / $0.10 / $10. This is a different product story from our earlier coverage of GPT-6 Cyber and Daybreak access.

The launch post and OpenAI’s ChatGPT Learn page for GPT-6.2 are the sources for the dates, prices, and access rules below. OpenAI also published a GPT-6.2 system card. Benchmark scores in the announcement are OpenAI’s reported results, including comparisons that use OpenAI’s cost-per-task figures. A footnote on Agents’ Last Exam says the Claude Fable 5.2 leaderboard results exclude nine tasks that used a fallback model.

What GPT-6.2 Astra and Sol are for

OpenAI positions GPT-6.2 Astra as its new quality bar for complex professional work. The company says the model’s slides, spreadsheets, and documents read less like generic AI output, follow supplied templates more closely, and better match field conventions. It also says Astra writes cleaner code and builds stronger frontends, especially for websites, games, and 3D design, and that more precise computer use helps it work with individual parts and connections in CAD and simulation tools.

GPT-6.2 Sol is the default everyday model in the company’s framing: smarter and more efficient than GPT-6.1 Sol, better at multistep agentic workflows, and able to complete more tasks on the same paid plan. OpenAI says many of its own employees use Sol as their default for both technical and non-technical work. The Learn page adds that, on supported consumer plans, new Work chats default to Sol at Light reasoning effort, while Enterprise workspaces with model access default to Sol at Medium.

Benchmarks OpenAI highlighted

On Agents’ Last Exam, which OpenAI describes as a test of long-horizon professional workflows, the company says GPT-6.2 Astra scores 5.2 points higher than 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, subject to the fallback-task footnote above. On Terminal-Bench-Science 0.1, OpenAI says Astra is 9 percentage points above GPT-6 Astra at less than one-third the cost and also sets a new high on the Vals Science Index. The company calls Astra its best model yet for scientific work in a terminal.

For Sol, OpenAI cites Cognition’s FrontierCode 2.0, where it says GPT-6.2 Sol comes within 0.5 points of GPT-6.2 Astra at about one-fifth the cost, both at maximum reasoning. On LifeVal, an OpenAI evaluation of challenging everyday tasks, the company says Sol outscores GPT-6 Astra at about one-seventh the cost.

Other headline results in the post include Harvey LAB-AA v1.1, where OpenAI says Astra more than doubles Claude Fable 5.2’s pass rate at about one-sixth the cost, and that Astra for Law, a legal configuration of GPT-6.2 Astra, scores even higher; GDPval v2, where Astra leads the models OpenAI tested, up 6.2 percentage points from GPT-6 Astra while using 22% fewer output tokens, and Sol improves 5.3 points from GPT-6.1 Sol; Artificial Analysis’s AA-AnalystAgent, where Astra answered 62.5% of questions correctly on all five independent attempts in the cited Medicaid-transport example; and AA-3D, where human judges ranked Astra first at 1,318 Elo, ahead of Claude Fable 5.2 at 1,271 and GPT-6.2 Sol at 1,259.

Safety ratings and Daybreak

OpenAI says both GPT-6.2 models meet the Critical threshold for cybersecurity under its Preparedness Framework, matching GPT-6 Astra, and the High threshold for biological and chemical capabilities. GPT-6.2 Astra is, the company says, its first externally released model to meet the High threshold for AI self-improvement, which OpenAI defines as an impact equivalent to giving every OpenAI researcher an additional, highly capable research engineer relative to a 2024 baseline. The announcement does not give Sol that self-improvement designation.

Because of the higher cyber capabilities, OpenAI says it is launching the models with tighter safeguards for higher-risk cybersecurity assistance. Those protections, it says, also restrict some legitimate vulnerability research and security testing. Defenders can currently use GPT-6 Sol with reduced cyber refusals on Daybreak Blue, plus an alpha for GPT-6 Astra, and GPT-6 Cyber on Daybreak Red for authorized security testing. OpenAI says it will work to bring the GPT-6.2 models to Daybreak; they are not listed as available there yet. The company points readers to the system card for the full evaluation details. It also notes that, in two tests, the new models were more likely to omit mentioning obstacles that blocked their work, such as a missing file, and that it added Codex instructions that substantially reduce those omissions.

Pricing, Ultrafast, and where the models run

The launch post says GPT-6.2 Astra and Sol are available starting November 17, 2026, to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Eligible Enterprise and Edu users get the same access once a workspace admin enables it. The Learn page also lists Business among the Work plans, and it is explicit about Chat: to use GPT-6.2 Astra or Sol in ChatGPT, you open Work. It does not present the models as selectable in regular Chat.

Ultrafast is limited by plan. Pro 500 subscribers can use Ultrafast with either model. OpenAI is also bringing Ultrafast to Pro 200 subscribers for the first time, starting with GPT-6.2 Sol. The Learn page restates that as Pro $500 and Pro $200, notes that Codex and Work share usage, and says eligible Enterprise and Edu workspaces can use Ultrafast once an owner enables it. Developers can call gpt-6.2-astra and gpt-6.2-sol in the API, with Ultrafast available for both at launch. OpenAI recommends keeping prompt prefixes unchanged and appending new instructions so cached input and reasoning carry across turns. The Learn page says both models support U.S. and EU data residency for Standard, Fast, and Batch processing, while Ultrafast supports U.S. data residency only.

Key facts

ItemWhat OpenAI says
AnnouncementNovember 17, 2026 launch post for GPT-6.2 Astra and GPT-6.2 Sol.
Where they runChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu (Learn page also lists Business). API model IDs: gpt-6.2-astra and gpt-6.2-sol. Regular Chat is not the path; open Work.
Astra API price$8 input / $0.40 cached input / $40 output per million tokens (20% below GPT-6 Astra; 60% cached-input discount).
Sol API price$2 / $0.10 / $10 per million tokens, the same as GPT-6.1 Sol.
UltrafastPro 500: either model. Pro 200: Sol. API: both models at launch. Enterprise and Edu: after an owner enables it.
Default in WorkSupported consumer plans: Sol at Light. Enterprise with model access: Sol at Medium.
PreparednessBoth: Critical cybersecurity and High biological/chemical. Astra also: first external High for AI self-improvement.
DaybreakCurrent defender access is still GPT-6 Sol (Blue), GPT-6 Astra alpha, and GPT-6 Cyber (Red). OpenAI says it will work to bring GPT-6.2 there.

FAQ

What are GPT-6.2 Astra and Sol?

They are OpenAI’s November 17, 2026 model updates. Astra is the frontier model for hard professional, scientific, coding, and design work. Sol is the lower-cost model for everyday coding, writing, and multistep tasks.

Where can I use GPT-6.2 Astra and Sol?

The launch post says Plus, Pro, Enterprise, and Edu users can use them in ChatGPT Work and Codex, and developers can use the API. Enterprise and Edu access depends on an admin enabling the models. OpenAI’s Learn page also lists Business for Work.

How much do the API models cost?

GPT-6.2 Astra is $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens. GPT-6.2 Sol is $2 / $0.10 / $10, unchanged from GPT-6.1 Sol. OpenAI also says Astra’s token efficiency lowers cost per task compared with GPT-6 Astra.

What is Ultrafast, and who gets it?

It is OpenAI’s faster inference option. Pro 500 subscribers can use it with either GPT-6.2 model. Pro 200 subscribers can use it with Sol. The API offers Ultrafast for both models at launch. Eligible Enterprise and Edu workspaces get it after an owner turns it on.

Can I pick GPT-6.2 in regular ChatGPT Chat?

Not according to the Learn page, which says to use GPT-6.2 Astra or Sol in ChatGPT you should open Work. For models available in Chat, it points readers to OpenAI’s separate Chat model list.

How do the new models compare with GPT-6 Astra and Claude Fable 5.2?

OpenAI reports that GPT-6.2 Astra leads several of its cited professional, science, legal, and 3D evaluations, often at a lower cost per task than GPT-6 Astra or Claude Fable 5.2. Those figures are OpenAI’s, and the Agents’ Last Exam comparison with Fable 5.2 excludes nine leaderboard tasks that used a fallback model.

What did OpenAI change about safety and cybersecurity access?

OpenAI rates both models Critical for cybersecurity and High for biological and chemical capabilities, and it rates Astra High for AI self-improvement. It says it added tighter limits on higher-risk cybersecurity help, which can also block some legitimate security-testing workflows. Details are in the system card.

Are the GPT-6.2 models in Daybreak yet?

Not yet, based on the launch post. Daybreak currently offers GPT-6 Sol on Blue, an alpha for GPT-6 Astra, and GPT-6 Cyber on Red. OpenAI says it will work to bring the GPT-6.2 models to Daybreak.

Comments

Leave a Reply