OpenAI Makes GPT-5.6 Luna the Free ChatGPT Default — With Unlimited Text Chats
Free and Go users get unlimited text conversations on GPT-5.6 Luna, which OpenAI says produces 62% fewer responses with factual errors than the prior default. Paid tiers get an improved GPT-5.6 Sol with a new reasoning-depth slider.
OpenAI has removed rate limits on text chats for free ChatGPT users and made GPT-5.6 Luna their default model, replacing GPT-5.5. The change rolls out to Free and Go tiers this week; caps remain only on file uploads, image generation, and other tools.
The quality claim attached to the swap is specific: responses containing at least one factual error are 62% less common with Luna than with the previous free-tier default, per OpenAI’s own factuality evaluations. Luna is also built for speed — it’s the fast-lane model, while a new “Think” button arriving next week lets free users invoke deeper reasoning on demand rather than being routed to it.
Paid users get their own upgrade. GPT-5.6 Sol, the unified model serving Plus and Pro subscribers, has been updated for reliability and more focused answers, and gains a slider that controls how much thinking effort ChatGPT applies to a response. That’s a notable interface shift: reasoning depth is becoming a user-facing dial instead of an invisible routing decision.
Free is the new battleground
Unlimited free access to a current-generation model is an aggressive move in a market that has been racing toward zero. DeepSeek’s V4-Flash undercuts Western APIs by roughly 99%. Google trimmed Gemini Flash pricing the same week. Meta is pushing its aggressively priced Muse Spark line. OpenAI’s answer is to stop metering conversations entirely at the bottom of the funnel and monetize tools, files, and multimodal features instead.
The strategic read: ChatGPT’s free tier is no longer a trial — it’s the product’s distribution engine. With weekly active users in the hundreds of millions, every rate-limit prompt was a churn point pushing users toward a free Gemini or Meta AI session. Removing limits trades inference cost (which GPT-5.6-class efficiency gains and OpenAI’s compute buildout have been driving down) for retention and data flywheel.
What it means in practice
If you’re a developer or a team standardizing on an assistant, the calculus shifts. The gap between free ChatGPT and a $20 Plus subscription now lives in tools — uploads, image generation, Sol’s deeper reasoning, and the thinking slider — not in whether you can keep chatting. Expect competitors to match unlimited text within months; nobody can afford to be the assistant that cuts you off mid-conversation while OpenAI doesn’t.
The unglamorous part matters most: the default free model got materially more accurate. Most of ChatGPT’s audience never touches a model picker. For hundreds of millions of people, “ChatGPT” just quietly became a 62%-less-wrong product — and that, more than any benchmark chart, is what moves trust at scale.