OpenAI Launches GPT-6.2 Astra and Sol: a Faster Frontier Model With a 20 Percent Cheaper API
OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, a new pair of AI models for professional and everyday work. Astra, the frontier model, is faster and cheaper than its predecessor, with API prices cut by 20 percent, while Sol keeps its price unchanged and nearly matches Astra on real-world coding tasks at about a fifth of the cost.
OpenAI Launches GPT-6.2 Astra and Sol: a Faster Frontier Model With a 20 Percent Cheaper API
OpenAI has released two new AI models, GPT-6.2 Astra and GPT-6.2 Sol, splitting its latest launch between demanding professional work and efficient everyday tasks. Both models arrived on 17 November 2026 in ChatGPT Work, the coding tool Codex, and the OpenAI API, and the headline change is price: Astra now costs 20 percent less than the model it replaces.
The release comes only weeks after the GPT-6 family itself, and it follows a pattern OpenAI has settled into this year — a frontier model for the hardest jobs, paired with a cheaper model that handles most ordinary work almost as well. What makes this pair different is how close the two now sit. According to OpenAI, Sol nearly matches Astra on real code changes to real codebases, at roughly a fifth of the cost.
Astra: built for the hardest projects
GPT-6.2 Astra is OpenAI's new frontier model, aimed at the work that strains lesser models: software development, scientific research, and long professional documents. OpenAI says it trained Astra to produce clearer, better-designed slides, spreadsheets, and documents that follow the conventions professionals use in fields such as law, sales, finance, and accounting, and that the difference is visible in slideshows, websites, and 3D designs.
The benchmark numbers OpenAI published are its own, and should be read as the company's claims rather than independent findings. On Agents' Last Exam, a test of multi-step professional work, Astra scored 5.2 points above GPT-6 Astra at less than half the cost per task. OpenAI also says Astra sets new highs across professional work and scientific research at less than a third of the cost per task of GPT-6 Astra.
The most striking claim concerns cybersecurity. OpenAI calls Astra its most capable model ever for cyber work, saying it jumps 22 points over GPT-6 Astra on ExploitBench Internal Port, an internal benchmark for exploit development. Because of that jump in capability, OpenAI is launching both models with tighter safeguards around higher-risk cybersecurity assistance, and says it is working to bring them to Daybreak, its programme that gives approved defenders access with reduced refusals.
Sol: the everyday model that nearly keeps up
GPT-6.2 Sol is the efficient half of the pair, designed for everyday coding, writing, and multi-step agent workflows. OpenAI describes it as a model that powers through routine work, writes clearly and naturally, and is intuitive to collaborate with — and says many of its own employees already use Sol as their go-to model across technical and non-technical work.
The comparison that matters most for users is the one OpenAI draws on FrontierCode 2.0, a benchmark built on merge-ready changes to real codebases. There, Sol nearly matches Astra at about one-fifth of the cost. Independent evaluator Artificial Analysis reached a similar conclusion from a different angle, scoring Astra at 62 on its Intelligence Index, up from 57 for GPT-6 Astra, and Sol at 57, up from 55 for GPT-6.1 Sol.
Both models offer a 1.05-million-token context window, with up to 922,000 input tokens and 128,000 output tokens per request, according to OpenAI's model pages.
Prices, plans, and where the models appear
The price story is simple, and it is the part of the launch OpenAI is pushing hardest. Astra's standard API rates are 8 dollars per million input tokens and 40 dollars per million output tokens, 20 percent below GPT-6 Astra, with cached input at 0.40 dollars per million tokens — a 60 percent cut. Sol keeps exactly the rates of GPT-6.1 Sol: 2 dollars input, 0.10 dollars cached input, and 10 dollars output per million tokens. At standard rates, Sol costs a quarter of Astra on every measure.
In ChatGPT, both models are available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. They are not being added to regular Chat, which keeps its existing models. Developers can call them through the API as gpt-6.2-astra and gpt-6.2-sol. Pro 500 subscribers can use Ultrafast, a faster generation mode, with either model; Pro 200 subscribers get Ultrafast for the first time, starting with Sol. The models are also generally available in GitHub Copilot on Pro+, Max, Business, and Enterprise plans.
OpenAI has also opened a background tier in its Responses API, aimed at work that can wait: choosing a two-hour or eight-hour completion window cuts token rates by 30 or 40 percent compared with standard processing.
A first for self-improvement risk
The safety picture is the part of this launch that will draw the most scrutiny. In the GPT-6.2 system card published alongside the release, Astra becomes the first externally released OpenAI model rated High for AI self-improvement under the company's Preparedness Framework — a measure of a model's ability to improve itself. Both models are rated Critical for cybersecurity and High for biological and chemical capabilities.
That rating does not mean the models are unsafe to use; it means OpenAI's own framework places them in a higher risk band than any model it has released before, and the tighter cybersecurity safeguards announced with the launch follow directly from it. The system card is the document to watch for how OpenAI justifies releasing a model at that level, and for what mitigations it says are in place.
What this means for users
For most people, the practical effect is straightforward. Paying ChatGPT users get two new models in Work and Codex without a price change to their plans, and developers get a frontier model that costs a fifth less than its predecessor, plus an everyday model that OpenAI itself says is close enough for most coding work at a quarter of the price.
The launch also keeps the pressure on OpenAI's rivals. Astra's new price of 8 and 40 dollars per million tokens matches the pricing of Claude Fable 5.2, released eight days earlier, and the contest between the two frontier models — on capability, on price, and on safety ratings — is now the defining race in the AI industry. The GPT-6 line's earlier cyber rating, which drew a Critical cybersecurity assessment for GPT-6 itself, is now the baseline Astra has to be measured against, and OpenAI's science work with GPT-6 Astra shows the direction the frontier model is being pushed. Meanwhile, ChatGPT itself keeps expanding into new territory, most recently with a wallet that lets AI agents shop and check out for users.
Whether Astra's benchmark gains hold up outside OpenAI's own tests will become clear as developers and researchers put the models through their paces in the coming weeks. For now, the release is live, the prices are lower, and the safety questions are sharper than they have ever been for a model OpenAI has put in public hands.
Sources
- OpenAI Community — GPT-6.2 Astra and Sol are now available (17 November 2026)
- Unite.AI — OpenAI Releases GPT-6.2 Astra With First High Self-Improvement Rating
- WinBuzzer — OpenAI Releases GPT-6.2 With Higher Performance and Speed at Lower Cost
- Ajay Kumar — OpenAI GPT-6.2 Astra and Sol: Prices, Limits, Background Windows
- The Times of India — OpenAI launches GPT-6.2 Astra and Sol artificial intelligence models
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Angry
0
Sad
0
Wow
0