GPT-5.6 Sol’s ChatGPT update is a surface change, not a Codex change

OpenAI said on August 6 that it had updated GPT-5.6 Sol inside ChatGPT for Plus and Pro users. The change pairs response-style tuning, aimed at more focused answers and tighter formatting, with a control that lets users set how much thought ChatGPT applies. OpenAI also says the same ChatGPT model now underlies both Instant and deeper-reasoning interactions, so changing effort should preserve its tone and behavior rather than switch the user to a separately characterized model.

The nearest comparison OpenAI provides is GPT-5.5 Instant, whose example response spends far more words reiterating a weather query; the company uses that contrast to argue that Sol should answer the central question first and shorten follow-ups. It also reports that, in its internal financial, medical and legal factual-detail evaluation, answers containing at least one factual error were 68% less common for Sol than for GPT-5.5 Instant. That is a vendor measurement, not an independently reproduced benchmark, so it substantiates OpenAI’s product rationale but cannot establish a general accuracy gain across deployments.

The scope matters more than the model label. OpenAI says the revised Sol is optimized for everyday chats and is only available in the Chat experience; it expressly says the Sol version used by ChatGPT Work and Codex is not changing in this update. The company had previously described Sol’s paid ChatGPT rollout as targeting complex work across coding, research, science, cybersecurity, computer use and design. The newer announcement therefore gives technical teams a reason to separate observations about ChatGPT’s interface behavior from assumptions about Codex, Work, or API behavior.

Practically, a user evaluating the update can test two narrow claims: whether the effort control produces a fuller answer without a sudden change in voice, and whether concise prompting is handled with less redundant formatting. Neither test confirms the internal error-rate result or substitutes for workload-specific evaluation. The update also does not remove all limits: OpenAI says free users’ expanded text access applies to GPT-5.6 Luna, while uploads, images and other tools retain limits. For engineering organizations, the safest reading is a ChatGPT product adjustment whose useful effect must be validated against their own prompts and operational constraints.