ChatGPT hits 1B weekly users as OpenAI drops text limits and cuts errors 68%

OpenAI is uncorking a major update that kills text chat limits for everyone and puts a sharper brain inside ChatGPT, a move that lands just as the service rockets past a billion weekly users. The new default engine, GPT-5.6 Luna, replaces the outgoing GPT-5.5 for Free and Go tiers. Alongside it comes a “Think” button, giving non-paying users their first toggle to crank up reasoning power on thorny questions—a feature that was previously locked behind a paywall. File, image, and voice generations still have their own guardrails, but pure text is effectively a free-for-all now.

Paying users aren’t getting a consolation prize; they’re getting a different car. Plus and Pro subscribers unlock GPT-5.6 Sol, a model that’s separate from the Codex and Work version and tuned for rapid-fire tasks: web research, planning, decisions, writing. OpenAI promises answers that are “more compact and robust.” More interestingly, they’re getting a thinking slider—not just an on/off switch, but a dimmer for cognition. You can dial up the reasoning steps based on how tangled the problem is, which introduces a degree of user-controlled resource allocation that feels genuinely new.

The numbers are what shift this from a routine update to a consequential one. Internal testing showed factual errors dropping 62% for Luna and 68% for Sol when measured against GPT-5.5-Instant. That’s not a marginal polish; it’s a meaningful reduction in the model’s tendency to confidently fabricate. Hallucination has been the persistent Achilles’ heel of large language models, and a two-thirds improvement puts Sol in a different reliability class for tasks where being wrong has actual costs.

Rolling out now for subscribers and hitting free users this week, the unlimited access and Think button go live next week. The timing is hard to ignore: a billion weekly users puts ChatGPT in a rarefied tier of internet services—think YouTube or Facebook scale—and removing text limits is a defensive moat as much as a feature. Competitors like Claude and Gemini have been chipping away with their own free-tier arguments, and OpenAI’s answer is to simply open the floodgates. The question that lingers is whether the infrastructure holds. Unlimited access for a billion users is a stress test, not a launch party, and any degradation in speed or uptime will be the real review of GPT-5.6.