GPT-5.6 Sol's latest change is a ChatGPT tuning pass, not a new developer-model rollout

OpenAI said on August 6 that it updated GPT-5.6 Sol in ChatGPT for Plus and Pro users. The stated changes are more focused answers, stronger factual reliability, and an effort slider that lets users choose how much thought ChatGPT applies. OpenAI also says the same ChatGPT model now underpins both Instant responses and deeper reasoning, aiming to make a higher-effort reply feel like a more thorough version of the same assistant rather than a switch to a different personality.

The timing matters because the change follows OpenAI's July 30 GPT-5.6 positioning, which emphasized a family split between the lower-cost Luna and Terra models and Sol for higher-end work. In that earlier announcement, OpenAI introduced Fast mode for API use of Sol, replacing Priority Processing and promising up to 2.5 times the Standard-processing speed at twice the price, without a claimed intelligence change. The August 6 post is narrower: it describes a ChatGPT interaction update, not a revised Sol API price or a new API capability.

OpenAI compares the revised ChatGPT behavior with GPT-5.5 Instant using an example that favors answering a rider's practical question before elaborating. It also reports an internal evaluation in financial, medical, and legal prompts where responses with at least one factual error were about 68% less common for GPT-5.6 Sol than for GPT-5.5 Instant. That number is useful as a vendor-reported directional result, but it is not an independent benchmark and does not establish the same improvement for a particular production workflow.

The company has separately shown that harness choices can materially alter its own results. In a July ARC-AGI-3 report, OpenAI said GPT-5.6 Sol scored 13.3% on the public set with the benchmark's official harness and 38.3% after retained reasoning and compaction were enabled. The report supports a practical caution for technical users: observed quality can depend on context retention, tool configuration, and effort settings, so a ChatGPT behavior update should not be treated as a general performance claim for every Sol deployment.

The clearest operational boundary is that OpenAI says this version of Sol is optimized for everyday ChatGPT conversations and is available only in the Chat experience; the Sol version used by Work and Codex is not changing in this update. Teams using the API, Codex, or ChatGPT Work therefore should not assume a new default model behavior from the August post. For ChatGPT users, the relevant change is controllability of answer depth and a stated attempt to reduce factual mistakes, while file uploads, images, and other tools still retain their own limits for free users.