OpenAI updates GPT-5.6 Sol, cuts ChatGPT factual errors by 68%

Plus and Pro subscribers get immediate access to the updated Sol model along with a new slider that lets them dial reasoning effort up or down for a given response, instead of the model deciding entirely on its own how much thought a question deserves. Sol sits alongside Terra and Luna as one of three GPT-5.6 variants, positioned specifically for everyday conversations. OpenAI describes the intent directly: the update is built to give more direct responses, tighter formatting, and less unnecessary detail when it doesn’t help the answer, with specific gains called out on dates, numbers, rules, sources, and assumptions.

That framing matters more for accuracy-sensitive questions than for casual ones: the 68% error reduction specifically covers financial, medical, and legal prompts, the categories where a wrong but confident answer causes the most damage, since a user has no easy way to catch a mistake buried in a fluent, confident-sounding response. A Think button sits alongside the slider for anyone who wants deeper reasoning on a single hard question without switching models entirely.

GPT-5.6 Luna, previously reserved for paid tiers, is becoming the default model for Free and Go accounts this week, giving anyone without a subscription access to a meaningfully stronger model than what free users had before this change, without any action needed on their end.

Free-tier text chats were previously capped, putting a hard ceiling on how much anyone without a subscription could actually use ChatGPT day to day. That cap is going away alongside the Luna upgrade, so free accounts get both a stronger model and unlimited conversation length in the same release, without needing to pay for either improvement separately. Image and file upload limits for free accounts stay exactly where they were before the update.