OpenAI locks free users out of GPT-5.6 Sol, cuts factual errors by 68% for paying tier

OpenAI just drew a sharper line between its paying customers and everyone else. The company is rolling out GPT-5.6 Sol to Plus and Pro subscribers, while free users get demoted to the cheaper, less capable GPT-5.6 Luna as their default model. The shift arrives alongside some genuinely impressive accuracy gains — and a new control that may just give users a different kind of headache.

The headline numbers are striking. OpenAI says factual errors dropped about 62 percent for Luna and 68 percent for Sol compared to the previous GPT-5.5 Instant, based on internal tests across finance, medicine, and law prompts. Those figures haven’t been independently verified, but if they hold up, it’s a meaningful step toward making these models trustworthy for professional work. GPT-5.6 Sol also cuts the fluff — OpenAI claims it now delivers direct answers for simple questions without burying you in bullet points, while still providing fuller responses for complex tasks.

Paying users get a new slider to control how much reasoning effort the model applies, with five settings ranging from quick answers to deep planning and coding. An OpenAI developer even published a breakdown of which setting fits which task. I’m skeptical this solves anything. Users largely ignored the old model switcher, and if you have to guess which of five reasoning levels to use for a given prompt, the interface has failed you. It’s a power-user feature dressed up as a usability improvement.

Free users, meanwhile, are losing access to OpenAI’s most capable reasoning. GPT-5.6 Luna takes over as the default later this week, with unlimited text chats and a “Think” button for harder questions arriving next week. But that button only lets Luna think longer — it doesn’t switch to a stronger model. Smaller models are inherently more error-prone, and we saw exactly that risk play out with the previous Instant model, which produced a completely wrong data analysis that its larger counterpart handled correctly. File uploads, image generation, and other tool limits stay in place too. The tradeoff is clear: free access gets broader but shallower.