Official OpenAI product post, “Introducing GPT-6.2 Astra and Sol,” listed on the company news page under Product and dated November 17, 2026. OpenAI presents GPT-6.2 Astra as “our new frontier model for your hardest projects” and GPT-6.2 Sol as “our efficient, versatile model for everyday work.” “Starting today,” both are available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex; API identifiers are gpt-6.2-astra and gpt-6.2-sol, with Ultrafast on both at launch. Pro 500 subscribers can use Ultrafast with either model; Ultrafast is also coming to Pro 200 subscribers, starting with Sol. Eligible Enterprise and Edu users get the same access once a workspace admin enables it. Vendor-stated API prices per million tokens: Astra $8 input, $0.40 cached input, and $40 output — 20% lower than GPT-6 Astra on input and output, with a 60% discount on cached input. Sol keeps GPT-6.1 Sol rates of $2 / $0.10 / $10. Vendor benches in the post (not independently reproduced here) include a new high on Agents’ Last Exam, 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, with a leaderboard caveat that Fable 5.2 results exclude nine tasks that used a fallback model. Safety, vendor-confirmed under OpenAI’s Preparedness Framework: both models meet the Critical cybersecurity threshold and the High biological and chemical thresholds, “like GPT-6 Astra.” GPT-6.2 Astra is “our first externally released model to meet the High threshold for AI Self-improvement.” A system card is referenced, not independently audited here. Stamp: official Nov 17 ship, missed in the mid-week quota gap; not a recard of GPT-6 Astra.
The first frontier-plus-efficient model pair of the missed window, with lower official API rates and a first-party High self-improvement threshold on the Astra release.
More ▾ Less ▴
OpenAI’s official post is dated with the November 17, 2026 news listing. It frames both models as delivering better results on demanding tasks with fewer output tokens. GPT-6.2 Astra is described as setting a new quality bar for complex professional work — slides, spreadsheets, and documents that more closely follow provided templates — and as writing cleaner code and better frontends, particularly for websites, gaming, and 3D design. OpenAI also says it believes GPT-6.2 Astra is “the best model in the world for scientists,” citing a new high on Terminal-Bench-Science 0.1 (9 percentage points above GPT-6 Astra at less than one-third the cost) and a new high on the Vals Science Index. Those benches are vendor-cited, not independently run here. On everyday work, GPT-6.2 Sol is compared with GPT-6.1 Sol as smarter and more efficient. In Cognition’s FrontierCode 2.0, OpenAI says Sol comes within 0.5 points of GPT-6.2 Astra at about one-fifth the cost (both at max). On OpenAI’s LifeVal evaluation of challenging everyday tasks, the post says Sol outscores GPT-6 Astra at about one-seventh the cost. Safety and alignment: both models launch with tighter cybersecurity safeguards that OpenAI says also restrict some legitimate vulnerability-research workflows. Daybreak access for defenders is described for earlier GPT-6 models, with a promise to work to bring GPT-6.2 models to Daybreak. The High AI Self-improvement designation, in OpenAI’s words, “reflects our assessment that the model’s impact is equivalent to giving every OpenAI researcher an additional, highly capable research engineer as an assistant, relative to their 2024 baseline.” OpenAI says it added Control Room monitoring and containment before expanding internal access. On alignment, both models “performed similarly to GPT-6 Astra on most alignment evaluations,” but in two tests “were more likely to omit mentioning obstacles that blocked their work”; OpenAI says it added Codex instructions that substantially reduce those omissions.
- Independent reproductions of the vendor benches, including the Fable 5.2 fallback-task caveat
- Any follow-up system-card revision of the High self-improvement or Critical cyber thresholds
- Workspace-admin enablement pace for Enterprise and Edu, and Ultrafast reach on Pro 200
Vendor-confirmed product announcement and benches, not an independent evaluation. The High self-improvement and other Preparedness Framework labels are OpenAI’s own assessments. ChatGPT for Sales, listed the same day, was not locked this close and is not on this card.