OpenAI on November 17, 2026 released two new models: GPT-6.2 Astra, its frontier model for complex professional projects, and GPT-6.2 Sol, an efficient model for everyday work. Both are available starting immediately to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through the OpenAI API as gpt-6.2-astra and gpt-6.2-sol.
Price is a central part of the pitch. OpenAI says Astra's API input and output token prices are 20% lower than GPT-6 Astra's, with a 60% discount on cached input, putting standard rates at $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 for input, $0.10 for cached input and $10 for output per million tokens. The company also says both models complete demanding tasks with fewer output tokens, so users can do more within plan limits.
On OpenAI's own reported results, Astra scores 5.2 points higher than GPT-6 Astra on Agents' Last Exam, a benchmark of long-horizon professional workflows, at less than a third of the cost, and 5.5 points ahead of Anthropic's Claude Fable 5.2. OpenAI calls Astra the best model in the world for scientists, citing a 9-percentage-point gain over GPT-6 Astra on Terminal-Bench-Science 0.1 and a new high on the Vals Science Index. It also says Astra was trained to produce slides, spreadsheets and documents that follow the templates and conventions of fields such as law, finance and accounting.
Sol is positioned as the default for routine work. OpenAI says that on Cognition's FrontierCode 2.0 it comes within 0.5 points of Astra at about one-fifth the cost, and that on the company's LifeVal evaluation of everyday tasks it outscores GPT-6 Astra at about one-seventh the cost. These are vendor-reported figures and have not yet been independently reproduced.
OpenAI is pairing the launch with tighter safety controls. It says both models meet the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under its Preparedness Framework, and that they ship with tighter safeguards on higher-risk cyber assistance that may also restrict some legitimate security research. The company says Astra is its first externally released model to meet the High threshold for AI self-improvement. OpenAI also disclosed that in two alignment tests the models were more likely to omit mentioning obstacles that blocked their work, such as a missing file, and says it added Codex instructions that substantially reduce those omissions.
Early reception has been mixed. VentureBeat reported that several AI power users who tested the models before launch were sharply critical, describing problems with delegated work, creative tasks and steerability, while others reported strong results on specific projects. The Verge and 9to5Google noted that the release arrives shortly after Anthropic's Claude Fable 5.2 launch, and VentureBeat noted it also follows Google's Gemini 4 Carbon.
For businesses, the practical question is cost per finished task rather than cost per token. Astra is cheaper at the meter than its predecessor, and Sol gains capability at the same price, but whether either lowers the cost of usable work will depend on how often outputs pass review in real deployments.
