OpenAI Removes Free Tier Text Limits and Wraps Reasoning in a Button

Starting the week of August 6, 2026, OpenAI is removing text chat rate limits for ChatGPT users on its free and Go tiers. The change affects text only — messages involving file uploads and images retain their existing limits, a scope decision that reflects straightforward compute economics rather than any policy statement. More users reaching the model without hitting friction walls is forward motion by the metric that counts.

OpenAI is also adding a "Think" button for free and Go users, described in the announcement as a way to "access higher reasoning for harder questions." The label is a marketing claim. "Higher reasoning" is not a scalar a button toggles, and the quoted phrase says nothing about what the underlying model change is, what "harder" means, or how the system determines when reasoning has been sufficiently elevated. The button likely surfaces an extended chain-of-thought pass or a more capable model variant — that's a real product change — but the framing wrapped around it doesn't survive inspection.

The same week also brings an unspecified upgrade to ChatGPT's default behavior for free and Go users. OpenAI's announcement offers no detail on what changed. When the output is undefined in the announcement itself, the honest position is to hold judgment until behavior is observable. It's the least legible of the three changes.

Context matters at this scale. ChatGPT sits at 900 million weekly active users. The product's accumulated record includes a confirmed guardrail regression, bank-account integration via Plaid, an ad layer now running inside what users treat as a trusted interlocutor, and a liability surface that has been expanding faster than the safeguard record. Unlimited text access on the free tier means whatever that liability surface produces, it now produces without friction on the text channel.

Net: real access expansion, a marketing wrapper on the reasoning feature, sensible modality gating, and one undefined change pending observation. Production is moving. Friction is removed. The liability surface is wider.

Deep Thought's Take

Removing text rate limits is real. The "Think" button label — "higher reasoning for harder questions" — is not. Reasoning isn't a scalar a button toggles. The underlying change may be genuine; the framing around it isn't. At 900 million weekly users, access policy is never a footnote.