OpenAI is removing the message ceiling from ordinary ChatGPT text conversations for Free and Go users. It sounds like a simple limit increase, but the more important change is architectural: OpenAI is separating high-volume everyday chat from the more expensive tools and reasoning modes around it.
The rollout makes GPT-5.6 Luna the default model for those two plans, replacing GPT-5.5. Users will also see a new Think button that can request more reasoning power for a difficult question.
That combination matters. Unlimited access keeps routine prompts moving, while the Think button gives users a deliberate way to spend more compute only when the task deserves it.
Unlimited text is not unlimited ChatGPT
OpenAI’s current ChatGPT documentation, the new policy applies to text chats. Files, image inputs, voice, and image generation will continue to have separate limits.
In other words, free users can stop rationing basic questions, drafts, summaries, and follow-up messages. They should not expect endless document analysis or image creation. Those features consume different infrastructure and remain useful boundaries between free and paid plans.
OpenAI is making conversation abundant, not making every expensive capability free.
That distinction is easy to miss in the word “unlimited.” The core chat loop is becoming a utility, while multimodal tools and heavier inference remain metered resources.
Luna is the key to the economics
Serving unrestricted chat to a product with more than 1 billion weekly users would be difficult if every message ran through the most computationally demanding model. GPT-5.6 Luna appears designed to solve that problem: it becomes the broad default, while additional reasoning is invoked explicitly.
This is a familiar systems pattern. Use the efficient model for the common path, reserve more compute for the smaller share of requests that need it, and expose the choice in the interface. The Think button is therefore more than a new control; it is a visible version of model routing.
OpenAI’s current ChatGPT documentation positions Chat as the everyday mode for asking questions, learning, brainstorming, comparing options, and planning. Removing the message meter from that core workflow could change user behavior more than adding another specialist feature. Longer conversations become practical, and people can refine an answer without wondering whether each follow-up is consuming a scarce allowance.
Paid plans shift toward quality and control
Plus and Pro subscribers are getting a different upgrade. They gain access to an improved GPT-5.6 Sol model aimed at quick questions, web research, planning, writing, advice, and decision support. OpenAI says this ChatGPT-specific version should produce more compact and robust answers.
It is separate from the GPT-5.6 Sol variants used in Codex and Work, which are not changing as part of this release. That detail is important for technical users: a shared model family name does not guarantee identical behavior across products, because system instructions, tool access, latency targets, and serving configurations can differ.
Paid users will also get a thinking slider for controlling how much reasoning the model applies. Free and Go users receive a simple Think button; subscribers receive a more granular control. The product segmentation is moving away from “can you chat at all?” and toward model quality, tools, modalities, and control over inference.
Accuracy claims still need real-world evidence
OpenAI’s internal evaluation found factual errors were 62% less common with GPT-5.6 Luna and 68% less common with GPT-5.6 Sol when compared with GPT-5.5 Instant, according to TechCrunch.
Those numbers are encouraging, but they are vendor-reported results rather than an independent benchmark. Error rates also depend heavily on the prompt, topic, tool availability, and how a factual mistake is defined. The useful test will be whether users notice fewer confident errors across long, ordinary conversations after the rollout reaches scale.
The updated Sol model is already arriving for Plus and Pro accounts. OpenAI plans to roll out Luna as the Free and Go default, followed by unlimited text chats and the Think button during the week of August 10.
For most free users, the practical takeaway is straightforward: normal text conversation should no longer require a message budget. The limits have not disappeared; they have moved to the parts of ChatGPT that cost more to run.