OpenAI is removing the message limits that have constrained ChatGPT's free tier since launch, giving users unlimited text conversations powered by GPT-5.6 Luna. The company is also adding a dedicated Think button that lets free and Go subscribers trigger deeper reasoning on demand, with both changes rolling out over the coming week.
The unlimited allowance applies to text chats only. Limits will continue to apply to file uploads, image generation and other tool-backed features, and OpenAI says the rollout is subject to safeguards against abuse.
What Changes for Free Users
Message caps have been the defining friction of ChatGPT's free experience. Users hit a ceiling mid-conversation and either waited out a cooldown window or upgraded, and that ceiling has functioned as the primary conversion mechanism for paid tiers since the product's earliest days. Removing it for text represents a substantial shift in how OpenAI is positioning the free product.
Alongside the cap removal, GPT-5.6 Luna becomes the default model for both Free and Go users this week. The upgrade is meaningful on accuracy: OpenAI reports factual errors dropped 62 percent for GPT-5.6 Luna and 68 percent for GPT-5.6 Sol relative to GPT-5.5-Instant, the model many free users had been served previously.
Reasoning on Demand
The Think button gives free users something they have not had before, which is explicit control over how much computation a query receives. Pressing it lets GPT-5.6 Luna spend additional time working through questions that benefit from extended reasoning, rather than defaulting to the fast response path optimized for typical conversational queries.
That control matters because the trade-off is real in both directions. Extended reasoning improves results on multi-step problems, mathematical work and code, but it adds latency that would be unwelcome on routine exchanges. Exposing the choice to users sidesteps the need for the system to guess correctly every time.
The Economics Behind It
Unlimited text chats for a free tier of ChatGPT's scale is a considerable inference commitment, and it is only plausible because serving costs have fallen sharply. Efficiency gains across model architecture, serving infrastructure and specialized inference hardware have compressed the per-token cost of frontier-class models to a fraction of what it was two years ago.
The competitive picture matters as well. Free tiers across the assistant market have grown steadily more generous, and message caps have become an increasingly conspicuous point of friction for a product competing on daily habit. Retaining users who might otherwise drift to a rival during a cooldown window has a clear value even before conversion is considered.
Where the Paywall Sits Now
The change relocates rather than removes the boundary between free and paid. With text conversation no longer metered, the remaining differentiators are the tool-backed capabilities that carry genuinely higher marginal costs: file analysis, image generation, longer context windows and access to the strongest reasoning models.
That is arguably a more durable structure. A paywall drawn around expensive capabilities rather than around message count aligns pricing with actual resource consumption, and it removes the most common source of user frustration without giving away the features that justify a subscription.
For users, the practical effect arrives next week. For competitors, the effect is a new baseline expectation for what a free AI assistant should offer.
