GPT-5.6 Sol's ChatGPT update separates chat tuning from developer workflows

OpenAI updated GPT-5.6 Sol in ChatGPT on August 6 with a narrower product goal than a new model-family announcement: more focused responses, improved factual reliability, and a slider that lets Plus and Pro users choose how much reasoning effort ChatGPT applies. OpenAI also says the same Sol model now powers both Instant responses and deeper reasoning in ChatGPT, aiming to make the transition between quick and deliberate interaction feel more consistent.

The timing matters because GPT-5.6 was previously presented as a family spanning ChatGPT, Codex, ChatGPT Work, and the API, with Sol positioned as the flagship tier and Terra and Luna serving lower-cost roles. The August update changes the ChatGPT experience rather than reintroducing that family. OpenAI's current help documentation describes Sol as the engine for Medium, High, and Extra High reasoning on eligible paid plans, while Sol Pro remains the Pro option for harder and longer-running work.

Compared with the earlier ChatGPT behavior, the practical change is product-level continuity: OpenAI says Instant and higher-effort interactions should now share a more consistent tone and behavior, with the reasoning slider determining how much thought is used. That is different from a claim that every GPT-5.6 endpoint has been retuned. OpenAI explicitly says the version optimized for everyday chats is confined to ChatGPT's Chat experience, while the Sol version serving Work and Codex is not changing in this update.

For technically informed users, that boundary is the most important operational detail. Observed changes in ChatGPT answer length, formatting, follow-up behavior, or factual handling should not be treated as evidence that API, Work, or Codex workloads have changed. Teams evaluating those surfaces should continue to test the product and reasoning setting they actually deploy; the cited materials establish a ChatGPT-specific change, not a cross-product behavioral guarantee.

OpenAI reports that, on an internal set of financial, medical, and legal prompts requiring factual detail, answers with at least one factual error were 68% less common for the updated Sol than for GPT-5.5 Instant. That is a useful directional signal about the company's target, but it is not an independently reproduced benchmark and does not establish error rates for a particular domain, prompt distribution, or tool-enabled workflow. The source does not provide a public methodology sufficient to turn that figure into a general reliability estimate.

The accompanying access change is related but distinct: Free and Go users receive GPT-5.6 Luna rather than Sol, with unlimited everyday text chats and a Think control for harder questions subject to safeguards and separate tool limits. OpenAI's help page also says Luna, not Sol, powers Think for those plans. The result is a clearer segmentation: paid ChatGPT users see a revised Sol chat experience, while free-tier access expands through Luna; neither point by itself indicates a change to the already available Sol API or Codex model.