OpenAI retunes GPT-5.6 Sol for ChatGPT, leaving Codex unchanged

OpenAI has updated the GPT-5.6 Sol experience in ChatGPT for Plus and Pro users, adding a slider that lets them choose how much reasoning an answer receives. The company describes the revision as more focused and more reliable for everyday conversations, research, planning, writing, coding, and decisions. Crucially, this is a ChatGPT product update: OpenAI says the GPT-5.6 Sol used in ChatGPT Work and Codex is not changing.

That distinction matters because GPT-5.6 did not arrive as one uniform product. During its initial limited preview, OpenAI positioned Sol as the flagship tier, Terra as the balanced tier, and Luna as the fast, low-cost tier, with staged availability across API, Codex, and ChatGPT. The current documentation still assigns Sol to paid-plan reasoning in standard ChatGPT, while Terra and Luna are available in Work, Codex, and the API. A report based on ChatGPT Sol behavior therefore should not be generalized to an API, Work, or Codex workload without testing that specific surface.

The clearest performance evidence OpenAI offers for the Chat update is an internal evaluation of financial, medical, and legal prompts: it says responses containing at least one factual error were about 68% less common for Sol than for GPT-5.5 Instant. That is a useful vendor signal, but it neither isolates this Chat-only revision from the broader GPT-5.6 model nor establishes outcomes for production coding or research workflows. Axios independently confirmed the paid and free product changes, while noting the competitive pressure that lower-cost open-weight models have put on frontier-model pricing; it did not provide an external accuracy audit.

For paid ChatGPT users, the practical change is a control surface rather than a migration path. Medium, High, and Extra High reasoning map to Sol where a plan makes those options available, and the new slider exposes the trade-off between a quick response and more deliberation. Free and Go users are a separate path: their default becomes Luna and their Think control uses Luna rather than Sol. Engineering teams using Codex should therefore treat the announcement as a reason to check Chat behavior, not as evidence that their coding-agent model changed.

There are material limits to the claims. Availability is rolling out by plan and can be controlled in managed workspaces, so eligible users may not see the same options at the same time. OpenAI also says higher-risk biology and cybersecurity requests can be refused or subjected to additional checks. Most importantly, the supplied sources do not include an independent evaluation of this revised ChatGPT Sol configuration. The sensible near-term test is task-specific: compare the same representative prompts at the chosen reasoning level, then evaluate answer quality, latency, and the frequency of follow-up corrections before changing a team workflow.