OpenAI has removed the message limit on text chat for free ChatGPT users. It also added a “think” button that activates extended reasoning from GPT-5.6 Sol on complex questions, and expanded access to GPT-5.6 Luna for non-paying users.

Why it matters to you

Until yesterday, if you recommended ChatGPT to a colleague without a subscription, they’d hit a wall after a few messages due to rate limiting. That barrier is gone. The adoption hurdle shifts from the number of turns to the quality of questions: which is where you want it. The “think” button brings step-by-step reasoning, previously restricted to paid tiers, into the free version. For anyone scaling with AI, this means the people you collaborate with now have access to the same reasoning capacity you use. Conversations that ended with “I’ve run out of messages” can become real work today.

If you want to try it

Open ChatGPT without a paid account, ask a question that requires multiple logical steps, and activate the think button to see the difference between direct response and extended reasoning.

In detail

Free access to GPT-5.6 for non-paying users was already underway. On August 2, OpenAI opened frontier models to 100,000 academic researchers with real credits and stated admission criteria (as we reported then). Now the move broadens: no selection, no structured program. Access is for anyone who opens ChatGPT without a subscription.

What came before. Free ChatGPT users hit a rate limit that cut off conversation after a variable number of messages. The ceiling changed with server load: during peak hours, a few exchanges were enough to see the wait button. Extended reasoning, the model’s ability to pause and think before responding, was restricted to paid tiers.

What’s changing. The “think” button moves that boundary. When you activate it on a complex question, the model halts direct generation and dedicates compute time to reason through intermediate steps, then produces the answer. For a professional, this means delegating tasks like contract analysis or action plan drafting to the model, getting transparent reasoning to work from. It’s the same mechanic that paid tiers have had for months, now served to those who don’t pay. Access to GPT-5.6 Luna, the more compact variant in the family, also expands for free users.

Competitive context. The move comes as Chinese open-weight models pressure the accessibility front. Alibaba’s Qwen3.8 Max has reached first place on Artificial Analysis’ agentic rankings, matching GPT-5.6 Sol on multi-step tasks (as we reported yesterday). DeepSeek V4-Flash matches GPT-5.6 Luna at lower cost per task. If competitors offer frontier capability at zero or near-zero cost, OpenAI’s response is to lower the barrier on its most direct consumer product. Convergence toward similar capability at lower cost forces large providers to recalculate the perceived value of subscriptions, pushing toward broader offerings to retain user volume.

What we don’t know fully. Announcements don’t say how truly unlimited the free “think” really is. Paid plans have had dynamic rate limits even on reasoning, and OpenAI hasn’t published numbers on how many times a free user can activate extended reasoning before being throttled. The limit could be on think activations, not base text chat. That detail makes the difference between full access and free trial, and needs verification through real use over time. Anyone integrating these models into daily workflows should watch service latency and reliability on the free tier: aspects that often take a back seat to feature announcements.