GPT-5.6 Sol gets a ChatGPT-specific tuning pass while API and Codex stay put

OpenAI says it has updated GPT-5.6 Sol inside ChatGPT for Plus and Pro users, aiming for more focused answers, fewer factual errors, and more consistent behavior between quick replies and deeper reasoning. The new control is a slider that sets how much thought ChatGPT applies, available on web, mobile, and desktop. OpenAI explicitly limits the change to ChatGPT: it says the Sol version powering Work and Codex is unchanged.

That boundary is important because GPT-5.6 has recently received other, separate changes. On July 30, OpenAI lowered prices for the Luna and Terra API models and replaced Priority Processing with Fast mode, which it says can make Sol up to 2.5 times faster at twice the standard price while leaving Sol pricing unchanged. The August 6 ChatGPT update is therefore not evidence of a new public API model or a Codex refresh. It is a product-specific behavioral retuning of the interactive ChatGPT experience.

OpenAI's central quality claim rests on an internal evaluation: it says factual errors in financial, medical, and legal prompts were 68% less common with Sol than with GPT-5.5 Instant. That is a useful directional claim, but the announcement does not provide independent validation or a task-level breakdown that would establish accuracy for a particular production workflow. Axios independently reported the broader product shift, including OpenAI's move to make Luna the default for free users, but it likewise reports the supplier's assertions rather than an external benchmark.

For technical users, the practical split is clear. ChatGPT users can test whether the new slider improves planning, research, writing, or coding conversations at a tolerable latency, while teams that use Codex or the API should continue to benchmark those surfaces separately. The update may make interactive work feel more coherent across effort levels, but OpenAI has not claimed a corresponding API or Codex capability change. Treat the advertised reliability gains as a reason to evaluate the revised chat product, not as a substitute for workload-specific checks and guardrails.