Published: 7 August 2026
OpenAI released an updated version of GPT-5.6 Sol in ChatGPT on 6 August 2026, tuned for shorter, more direct answers and better factual reliability. In OpenAI’s own evaluation, responses containing at least one factual error were about 68% less common with the updated GPT-5.6 Sol than with GPT-5.5 Instant. OpenAI also made GPT-5.6 Luna the default model for Free and Go users this week, with unlimited text chats following next week.
What changed in GPT-5.6 Sol in ChatGPT?
GPT-5.6 Sol is the flagship model in OpenAI’s GPT-5.6 family, above the mid-tier GPT-5.6 Terra and the smaller, faster GPT-5.6 Luna. The 6 August update changes how that model behaves inside ChatGPT rather than what it can do in the API. OpenAI says the model now gives more direct responses, adapts the level of detail to the question, uses tighter formatting, and pushes back with a correction where simply agreeing would not be useful.
For Plus and Pro users, the same GPT-5.6 Sol now powers both Instant responses and deeper reasoning. Before this update, moving from a quick answer to a longer one meant switching to a model with a different tone and style. OpenAI’s stated goal is that raising the effort level should feel like the same model taking more time, not like handing the question to a different system.
The company frames this against scale: it says 1 billion people use ChatGPT each week for everything from quick questions and web searches to planning, research and complex decisions.
GPT-5.6 Sol accuracy: what the 68% figure actually measures
The headline accuracy claim comes from an internal OpenAI evaluation, not an independent benchmark. OpenAI tested financial, medical and legal prompts that require factual detail, and compared how often a response contained at least one factual error. Against GPT-5.5 Instant as the baseline, such responses were about 62% less common with GPT-5.6 Luna and about 68% less common with the updated GPT-5.6 Sol.
OpenAI has not published the prompt set, the sample size, or the grading procedure behind those two numbers, and no third party has reproduced them. The relative improvement is also measured against GPT-5.5 Instant, an older model, rather than against the previous version of GPT-5.6 Sol, so the figure does not tell you how much better the updated GPT-5.6 Sol is than the GPT-5.6 Sol that shipped in July. OpenAI attributes the gain to the model making better use of the sources it retrieves, especially where an answer depends on dates, numbers, sources, rules or assumptions.
Further detail on safety training sits in the accompanying system card, which covers additional training for users OpenAI believes to be under 18. For those users the model is trained to avoid romantic roleplay, age-restricted challenges and presenting itself as a substitute for real-world relationships, with age-appropriate boundaries around sexual content, eating disorders and body-image risks, age-restricted goods, dangerous activities and graphic violence.
What is the new ChatGPT reasoning slider?
Plus and Pro users get a slider that sets how much effort ChatGPT puts into each response. Lower settings are meant for everyday questions; higher settings are meant for planning, research, writing, coding and decisions that need more thought. The slider works on web, mobile and desktop, and the same control already existed in ChatGPT Work.
The design question is whether an effort control changes outcomes in practice. Model pickers in ChatGPT have historically been ignored by most users, and choosing the right model has generally mattered more than choosing a reasoning depth. The slider does make the trade-off visible rather than leaving it to a router, but it also adds one more setting to get wrong.
What do free ChatGPT users get with GPT-5.6 Luna?
GPT-5.6 Luna becomes the default model for Free and Go users this week. Starting next week, those users also get unlimited text chats, subject to abuse guardrails, plus a Think button that gives GPT-5.6 Luna more time to work through harder questions. Limits still apply to file uploads, image generation and other tools.
The important caveat is what the Think button does not do. It gives the smaller GPT-5.6 Luna more time; it does not route the question to GPT-5.6 Sol or GPT-5.6 Terra. Free users therefore get more volume and a longer-thinking small model, but no access to OpenAI’s stronger reasoning models. Smaller models are generally more error-prone on tasks that depend on precise retrieval or multi-step analysis, which is exactly the category the Think button invites people to use it for.
Where the update applies, and where it does not
This version of GPT-5.6 Sol is optimised for everyday conversation, so it only ships in the Chat experience inside ChatGPT. The version of GPT-5.6 Sol that powers ChatGPT Work and Codex is unchanged by this release, and OpenAI has not announced a matching API update. Teams that build on the GPT-5.6 Sol API endpoint should not expect the new answer style or the accuracy figures to carry over.
Full details can be found in OpenAI’s announcement on ChatGPT-5.6 Sol & Luna.
