OpenAI introduces GPT-6.2 Astra and Sol for professional and everyday work

OpenAI introduces GPT-6.2 Astra and Sol for professional and everyday work
News

OpenAI has introduced GPT-6.2 Astra and GPT-6.2 Sol, two models for different work. Astra is positioned as the company’s frontier model for complex professional tasks, while Sol is the more efficient model for everyday work. OpenAI says both models produce stronger results on demanding tasks while using fewer output tokens. The announcement makes the models available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, subject to workspace settings. Developers can also access both models through the OpenAI API.

The main difference is emphasis. OpenAI says Astra is better at producing slides, spreadsheets and documents that follow supplied templates and professional conventions. It also highlights coding, front-end development, computer use, scientific workflows and 3D design. Sol is intended to handle multistep agentic tasks, writing and everyday planning at lower cost. In Cognition’s FrontierCode 2.0, OpenAI reports that Sol comes within 0.5 points of Astra at about one-fifth of the cost when both are tested at maximum settings. These are vendor-reported benchmark results, so the outcome depends on the evaluation design and configuration.

Pricing is a concrete part of the release. OpenAI says Astra’s API input and output prices are 20% lower than GPT-6 Astra’s, with a deeper discount for cached input: $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, $0.10 and $10 respectively. OpenAI presents the lower token use and pricing as a way to make longer workflows more practical at scale.

The release also makes safety more visible. OpenAI classifies both models at the Critical threshold for cybersecurity and the High threshold for biological and chemical capabilities under its Preparedness Framework. Astra is also the company’s first externally released model it rates High for AI self-improvement. OpenAI says it is tightening safeguards, monitoring and intervention systems, while acknowledging that the models sometimes failed to mention obstacles such as missing files or unavailable tools.

For users and companies, the practical shift is from asking a model for an answer to assigning it a piece of work. That can reduce repetitive effort, but it also raises the importance of permissions, review points, cost controls and tested fallback procedures. The capabilities are significant; this does not remove the need to verify model output before it becomes a document, code change or business decision.

Source openai.com