GPT-5.6 Sol's latest change is a ChatGPT behavior tune, not an API revision

OpenAI announced on August 6 that Plus and Pro users are receiving an updated GPT-5.6 Sol in ChatGPT. The company describes the revision as a behavior and presentation tune: more focused answers, tighter formatting, adaptive detail, and a correction when agreement would be unhelpful. It also adds a slider on web, mobile, and desktop that lets those users choose how much thought ChatGPT puts into a response. This is a product update to an existing model experience, not evidence of a new model release.

The most concrete quality claim is OpenAI's internal evaluation of factual prompts in financial, medical, and legal domains. OpenAI says responses with at least one factual error were 68% less common for GPT-5.6 Sol than for GPT-5.5 Instant, with the corresponding reduction for GPT-5.6 Luna reported as 62%. That is useful directional evidence about the goal of the update, but it is vendor-reported evaluation data rather than an independent benchmark, so it does not by itself establish performance across other tasks or deployments.

The change is distinct from the July 30 API performance update. In that earlier announcement, OpenAI said Fast mode replaced Priority Processing for GPT-5.6 Sol, promising up to 2.5 times Standard-processing speed at twice the price and no claimed intelligence change; existing API requests marked priority would continue to work. The August 6 update instead changes ChatGPT interaction behavior and exposes an effort slider, so teams should not treat the newer consumer announcement as a reason to alter API latency, pricing, or routing assumptions.

OpenAI draws the boundary explicitly: the updated Sol is available only in ChatGPT's Chat experience, while the version powering ChatGPT Work and Codex is not changing in this release. That separation matters for technical users because a cleaner or more concise ChatGPT answer is not a documented behavioral change for a Codex workflow, an API integration, or a Work deployment. The sensible comparison is therefore between ChatGPT's prior Instant and Thinking experiences, rather than between this update and a competing developer model.

For ChatGPT users, the practical new control is the ability to trade response speed for more deliberation without changing the named model, while free users are being moved toward GPT-5.6 Luna with a separate Think button for harder questions. For engineering teams, the limitation is equally clear: OpenAI has not supplied independent measurements for the ChatGPT tuning, and it says Sol's Work and Codex version is unchanged. Evaluation on representative prompts remains the appropriate test before inferring gains in coding, research, or other production use cases.