On August 6, OpenAI shipped its biggest ChatGPT product change of the year so far: GPT-5.6 Sol gets retuned for paying users, while the default model for free users switches to GPT-5.6 Luna with unlimited text chats. The announcement came three and a half weeks after OpenAI cut API prices for the GPT-5.6 family, and it applies to what OpenAI says is now one billion weekly ChatGPT users. This change lands on both ends of the willingness-to-pay spectrum at once — the people paying twenty dollars a month and the people paying nothing get different, but both improved, experiences.
One Model from Instant to Deep Thinking
Until now, ChatGPT’s instant answers and deeper reasoning felt like two different products: different tone, different behavior, a small jolt of disorientation every time a question crossed the threshold. This update folds both experiences for Plus and Pro users into a single GPT-5.6 Sol. The official framing: it should feel like the model is taking extra time to think, not like you switched to a different model with its own personality.
The visible change is a slider. On web, mobile, and desktop, you choose how much thought ChatGPT puts into a single response — keep it low for everyday questions, push it up for planning, research, writing, code, or decisions that deserve more care. Sol itself is tuned to be more restrained. It adapts the level of detail to the question, avoids formatting nobody asked for, and offers a correction when simply agreeing with you would not be useful.
Fewer Factual Errors: 62% and 68%
Reliability is the part OpenAI emphasizes most, and the only hard numbers in the announcement. In an internal evaluation over financial, medical, and legal prompts that require factual detail, responses containing at least one factual error were about 62% less common with Luna and 68% less common with Sol than with GPT-5.5 Instant, the previous default. The improvement is attributed to better use of the sources the model retrieves when answering.
For regular users the meaning is plain: fewer confident wrong answers. For people who treat ChatGPT as a productivity tool, this matters more than any benchmark. The three categories OpenAI tested — finance, medicine, law — are exactly where a hallucinated number or a made-up deadline does real damage, and exactly where casual users increasingly go for advice.
Free and Go Tiers: Luna by Default, Unlimited Text, a Think Button
Starting this week, free users get GPT-5.6 Luna as their default model — a small, speed-focused model that only debuted in early July — with no cap on text chats. The low-cost Go subscription gets the same treatment. When a question needs more depth, a new Think button unlocks higher reasoning for harder problems, so the free tier keeps an escape hatch to the stronger end of the stack.
Axios cut to the chase in its coverage: the free plans from frontier labs now offer more intelligence than most people will use. What actually differentiates paid tiers is shrinking down to speed, headroom, and access to the strongest reasoning — not whether the thing works at all.
Why the Generosity, Why Now
The timing is easy to read. On July 30, OpenAI cut API prices for Luna and Terra, and Axios attributes the pressure to cheap Chinese open-weight models, which keep forcing OpenAI and Anthropic to prove their closed models justify a premium. Handing out a speed model like Luna without limits is a controlled cost: the model is cheap to serve by design, and unlimited text is the cheapest modality to give away. In exchange, the pitch for Plus moves from “it works” to “it works better” — sharper answers, a consistent voice, and one slider that scales effort instead of switching personas.
Two takeaways for product teams. First, a single model with adjustable thinking effort is becoming the standard interface for consumer AI — the era of shipping two models with two personalities inside one app is winding down, and users will expect effort to be a setting, not a product boundary. Second, free tiers are no longer a crippled-version warning; they are a customer acquisition funnel. Design on the assumption that your users can always get “just good enough” intelligence for free, and that whatever you charge for has to beat free in a way they can feel in a single session.
Sources
- Improving GPT‑5.6 Sol in ChatGPT—and expanding access to GPT‑5.6 Luna for free users — OpenAI
- OpenAI makes major upgrades for free and paid ChatGPT users — Axios
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.