GPT-5.6 Sol's ChatGPT update separates product behavior from the API and Codex model

OpenAI says its August 6 update changes GPT-5.6 Sol's behavior inside ChatGPT for Plus and Pro users: answers are intended to be more focused and more reliable with facts, and a slider lets users choose how much reasoning to apply. The company also says the same model now serves both Instant responses and deeper reasoning in Chat. Crucially, OpenAI describes this as a Chat experience update, not a change to the GPT-5.6 Sol deployment that powers Work and Codex.

The timing makes this a post-launch product adjustment rather than a new model-family event. OpenAI introduced GPT-5.6 Sol, Terra, and Luna across ChatGPT, Codex, and its API on July 9, with Sol positioned as the flagship tier. Its current help documentation says GPT-5.6 availability is still rolling out by eligible plan and workspace configuration. That means a user's observed behavior can depend on both the product surface and the account they are using, even when the family name is the same.

OpenAI's evidence for the Chat tuning is narrower than the broad capability case it made at launch. The August post reports an internal evaluation in which responses with at least one factual error were 68% less common for GPT-5.6 Sol than GPT-5.5 Instant on financial, medical, and legal prompts requiring factual detail. Its July launch material separately reports benchmark comparisons against GPT-5.5 and competing models. Those are useful provider disclosures, but they describe different evaluations and should not be read as a guarantee that the Chat behavior update improves every coding or research workflow.

For technical users, the practical consequence is to test the surface that will actually be used. The API documentation identifies `gpt-5.6` as an alias for Sol and lists its supported reasoning settings, tools, context window, pricing, and snapshots; the August Chat update does not claim to alter those interfaces. In ChatGPT, the new slider and eligibility rules matter, while Free and Go users receive Luna rather than Sol and managed-workspace administrators can control access. OpenAI also notes that safety checks can affect legitimate higher-risk requests, so repeatable evaluation should include availability, effort setting, tool use, and refusal behavior rather than comparing model names alone.