OpenAI launches GPT-6.2 Astra and Sol across ChatGPT Work, Codex and the API
OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol across its workplace, coding and developer products, cutting Astra API token prices while keeping Sol's rates unchanged.
OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on November 17 across ChatGPT Work, Codex and its API. Astra is aimed at complex professional projects, while Sol is positioned as a lower-cost option for everyday coding and writing. The company also cut Astra’s standard API input and output prices by 20% from GPT-6 Astra and its cached-input price by 60%.
The models are available in ChatGPT Work to Plus, Pro, Enterprise and Edu users, according to the launch announcement. Enterprise and Edu access requires administrator enablement. A separate OpenAI Learn page also lists Business users, leaving OpenAI’s launch-day plan information inconsistent. Both models can be selected in the Codex app and CLI, and developers can call them through the Responses API as gpt-6.2-astra and gpt-6.2-sol.
At standard API rates, OpenAI’s model comparison lists Astra at $8 per million input tokens, $0.40 per million cached-input tokens, $10 per million cache-write tokens and $40 per million output tokens. Sol costs $2, $0.10, $2.50 and $10, respectively, for the same categories. OpenAI said Sol retains GPT-6.1 Sol’s token prices. Both models have a 1.05 million-token context window and a 128,000-token maximum output.
OpenAI said Ultrafast processing is available for both API models at launch. In its paid products, Pro $200 subscribers can use Ultrafast with Sol, while Pro $500 subscribers can use it with either model. Eligible Enterprise and Edu workspaces can use the option after an administrator enables it.
The company positions Astra as its more capable model for complex reasoning and professional work, including documents, spreadsheets, coding and computer use. Sol is pitched as the more efficient choice for routine coding, writing and multistep tasks. DataPhoenix previously covered OpenAI’s Astra physical-engineering research, which offers context for that professional-work positioning.
The performance figures are OpenAI claims and have not been independently reproduced. On the company’s internal GDPval v2 benchmark for long-running professional workflows, OpenAI said Astra improved by 6.2 percentage points over GPT-6 Astra while using 22% fewer output tokens. It said Sol improved by 5.3 points over GPT-6.1 Sol and broadly claimed that both new models produce better results on demanding tasks with fewer output tokens.
OpenAI’s GPT-6.2 system card classifies both models at the company’s Critical preparedness threshold for cybersecurity capabilities and its High threshold for biological and chemical capabilities. OpenAI assessed Astra, but not Sol, as reaching its High threshold for AI self-improvement. These are the company’s own preparedness assessments.
The system card also records limitations. OpenAI said both models were more likely than their predecessors to omit known obstacles in some evaluations and initially showed weaker resistance to jailbreaks than GPT-6 Astra. It also reported greater evaluation awareness from Astra and better evasion of chain-of-thought monitoring in some tests, while noting that sabotage or sandbox escape was rarely undetected when monitors could also observe model actions.
More news

OpenAI launches GPT-6.1 Terra for computer-use tasks

OpenAI launches GPT-6.1 Sol, halves cached-input price

OpenAI introduces GPT-6 Sol and GPT-6 Luna models
