GPT-6.2 splits into Astra and Sol: more tasks on the same paid plan, with Ultrafast at launch
Synopsis
GPT-6.2 arrives as two models, Astra and Sol, and lands at once in ChatGPT Work, Codex and the Responses API. Astra targets complex reasoning across multiple sources, while Sol handles everyday coding, writing and multi-step work; compared with GPT-6.1 Sol, Sol lets users complete more tasks on the same paid plan. Both models support Ultrafast at launch in the Responses API, but that speed tier covers US data residency only.
Interpretation
GPT-6.2 no longer appears as a single model but as two selectable models, Astra and Sol, reaching three entry points: ChatGPT Work, the Codex app and CLI, and the Responses API. GPT-6.1 Sol is the baseline users compare against, and the new release hands model choice to users by task type. Compared with GPT-6.1 Sol, Sol lets users complete more tasks on the same paid plan, a direct comparison supplied by the releasing organization.
The two models divide work by task complexity: Astra handles complex reasoning that works across multiple sources and checks conclusions against the data, while Sol handles everyday coding, writing and work that takes several steps. Astra is positioned for complex reasoning and Sol for everyday multi-step tasks, called in the Responses API as gpt-6.2-astra and gpt-6.2-sol respectively. The stated use cases include giving Astra a failing workflow to trace a bug across services, implement a fix and run the affected tests, and asking Sol to review a pull request, revise code from reviewer feedback and run the relevant tests.
In standard, single-agent mode in the Responses API, users can change reasoning effort or switch from plain text to structured output between replies. The mechanism is to append a configuration_update item before the next user message while keeping request-level settings unchanged. Keeping earlier input intact helps preserve reasoning across turns and keeps that input eligible for prompt caching.
Perspective
The report covers what the release notes state: availability of the two models in Work, Codex and the Responses API, default models and reasoning effort by plan and workspace setting, subscription and regional conditions for Ultrafast, and the configuration-update mechanism in the Responses API. Benchmarks and safety evaluations are not detailed here and require the release announcement.
Ultrafast supports US data residency only, while Standard, Fast and Batch cover both US and EU regions; Enterprise and Edu workspaces require an admin to enable the models in workspace settings first, and Ultrafast also requires an owner to enable it. Pro $200 subscribers can use Ultrafast with Sol only, while Pro $500 subscribers can use it with either model.
