GPT-6.2 Astra / GPT-6.2 Sol
Intro
GPT-6.2 Astra and GPT-6.2 Sol are the pair of models OpenAI officially introduced on 17 November 2026. The post opens by calling Astra “our new frontier model for your hardest projects” and Sol “our efficient, versatile model for everyday work.” Both are described as delivering better results on demanding tasks with fewer output tokens, so more work fits within usage limits and costs make more sense at scale.
OpenAI says that starting that day, all Plus, Pro, Enterprise, and Edu users can use both models in ChatGPT Work and Codex; Pro 500 subscribers can use Ultrafast with either model, and Ultrafast is also being brought to the Pro 200 tier for the first time (check the official post for the exact plan details). The developer guide Using GPT-6.2 gives the API IDs gpt-6.2-astra and gpt-6.2-sol. OpenRouter’s catalog pages for GPT-6.2 Astra and GPT-6.2 Sol mark them as released on 17 November 2026, and describe Astra Pro as the same underlying model served with reasoning.mode set to pro, not a separate brand.
This slot’s incremental crawl of the openai.com sitemap newly listed the announcement page and an Astra for Law update. That page says Astra for Law is a model configuration of GPT-6.2 Astra (an expanded legal search index plus research, weighing, and writing instructions). The Astra for Law name itself was first introduced in September 2026 to a small group of firms, so this radar files only GPT-6.2 Astra / GPT-6.2 Sol as the new names.
Advertisement
Explain
OpenAI positions Astra as a new quality bar for complex professional work: slides, spreadsheets, and documents that more closely follow the templates you provide and the conventions of the field, plus cleaner code and stronger frontends, especially for websites, gaming, and 3D design. Benchmarks cited in the post include Agents’ Last Exam, where 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 about 5% of the cost (the post notes that Fable 5.2’s leaderboard result excludes nine tasks that used a fallback model); Terminal-Bench-Science 0.1, where Astra is 9 percentage points above GPT-6 Astra at less than one-third the cost; and a new high on the Vals Science Index. The post also cites internal GDPval v2, Harvey LAB-AA, and Artificial Analysis’s AA-AnalystAgent among other knowledge-work evaluations.
Sol is described as a go-to everyday model—“Many employees at OpenAI use it as their default model”—that is smarter and more efficient than GPT-6.1 Sol and lets users complete more tasks on the same paid plan. It is aimed at multistep agentic workflows and clear, natural writing. On Cognition’s FrontierCode 2.0, GPT-6.2 Sol comes within about 0.5 points of GPT-6.2 Astra at roughly one-fifth the cost (both at max). On LifeVal, an evaluation of challenging everyday tasks, Sol outscores GPT-6 Astra at about one-seventh the cost.
On API pricing, OpenAI says GPT-6.2 Astra’s input and output token prices are 20% lower than GPT-6 Astra’s, with a 60% discount on cached input: $8 input, $0.40 cached input, and $40 output, all per million tokens. GPT-6.2 Sol keeps GPT-6.1 Sol’s API token prices. OpenRouter lists Sol at $2 / $10 per million tokens with a 1.1M context window (confirm against the official and provider pages at the time you use them). The developer guide also warns that, starting with GPT-6.2, the API may drop reasoning items when it detects mid-conversation prefix edits, a model downgrade, or a change of end user or organization; tool calling should use the Responses API; and migrations should revisit reasoning.effort, unsupported sampling parameters, and the new prompt-cache options. OpenAI published a GPT-6.2 system card and states that both models meet the High threshold for biological and chemical capabilities under its Preparedness Framework, and that Astra is its first externally released model to meet the High threshold for AI Self-improvement, accompanied by tighter safeguards for higher-risk cybersecurity assistance.
Usage
In ChatGPT, OpenAI says Plus, Pro, Enterprise, and Edu users can choose GPT-6.2 Astra or GPT-6.2 Sol in ChatGPT Work and Codex. If a team already has Astra for Law, the updated configuration is selected as gpt-6.2-astra-law in ChatGPT, and Enterprise admins can request access through the Trusted Access Program. Those are entry points for the legal configuration, not the default for Astra / Sol themselves.
In the API, set model to gpt-6.2-astra or gpt-6.2-sol. OpenAI’s migration notes suggest starting with medium reasoning effort; if a prior integration used none or minimal, use low instead. Use reasoning.effort in the Responses API and reasoning_effort in Chat Completions. Both models support Chat Completions, but tool calling requires Responses. When migrating Chat Completions requests, omit temperature, top_p, logprobs, and top_logprobs. To improve cache reuse, lower cost and latency, and preserve reasoning across turns, OpenAI recommends keeping prompt prefixes unchanged and appending new instructions. The docs also state that both models support US and EU data residency with Standard, Fast, and Batch processing, that Ultrafast supports US data residency only, and that approved organizations can use Zero Data Retention on eligible endpoints.
For the full set of benchmark figures, examples, and safety notes, see the official announcement, Using GPT-6.2, and the GPT-6.2 system card linked from that post. Third-party routing (for example OpenRouter’s openai/gpt-6.2-astra and openai/gpt-6.2-sol, plus Astra Pro with pro reasoning mode) should be checked against each provider’s catalog and pricing at the time you use it.