Picture the moment, sometime in the week of August 10, when a free ChatGPT user typed a follow-up question and nothing stopped them. No countdown. No "you have used your messages" wall. No demotion to a smaller model. OpenAI removed the text message limit for free ChatGPT accounts, replacing the old 10-messages-per-5-hours cap with unlimited conversations powered by GPT-5.6 Luna. Announced on August 6 and rolling out across that following week, GPT-5.6 Luna became the default model for Free and Go users, and those users got unlimited text chats. For a service that OpenAI says reaches 1 billion people weekly, the implications land well outside any product changelog.
The move is not a single act of generosity; it is a carefully structured escalation. GPT-5.6 spans three model tiers: Sol, the flagship; Terra, a lower-cost model with performance competitive with GPT-5.5; and Luna, the fastest and most affordable model. Free users get Luna. The announcement also ships an updated GPT-5.6 Sol for Plus and Pro subscribers, retuned for everyday conversation; for paid users, the updated Sol is built to give more direct responses, use tighter formatting, and correct the user when simple agreement would not help. For Plus and Pro users, OpenAI also added a new slider in ChatGPT on web, mobile, and desktop that lets users determine how much time ChatGPT spends thinking about a response. Then, three days after the free-tier announcement, OpenAI previewed GPT-5.6 Sol Ultrafast on August 13, powered by Cerebras and claimed to run up to 14 times faster than standard processing, reaching up to 750 output tokens per second. The architecture of the whole product is moving: free users get breadth, paid users get depth and speed, and the gap between the two is designed to be just wide enough to motivate an upgrade.
The competitive logic is blunt. With ChatGPT now at 1 billion weekly users, the move is a direct push to hold that audience against Claude and Gemini, both of which already offer unlimited free text. Free Gemini wins on multimodal and search-grounded answers; free Claude is best for writing and careful reasoning, but has tighter limits. Until August, ChatGPT's free tier was arguably the most constrained of the three big players on raw message volume. That asymmetry is now inverted for text, the most common use case by a wide margin. The economics behind the pivot are visible too. Advertising revenue from the free and Go tiers is the financial mechanism that enables those tiers to remain viable; ad agency WPP's Kate Scott-Dawkins estimated OpenAI could earn roughly $500 million to $800 million in its first year of advertising. As of August 11, ChatGPT Ads had already launched in the United Kingdom, Mexico, Brazil, Japan, and South Korea. The bargain being struck with free users is now explicit: unlimited conversation, in exchange for ads below answers and, by default, training data.
That last clause deserves a hard look. By default, OpenAI enrolls all free accounts in model training. Free users may also see ads when using ChatGPT; they appear in a "sponsored" section below the chatbot's main response, and often take the form of products or services related to the prompt. OpenAI claims ads are contextually targeted based on the current query, not by selling chat history to third parties. Skeptics note that similar assurances from other platforms have evolved over time, and that opt-out, rather than opt-in, remains the default for data use. OpenAI claims Luna produces 62 to 68 percent fewer factual errors than GPT-5.5, but that figure comes from internal evaluation with no published methodology or third-party validation. The company is asking a billion people to trust both the model's accuracy claims and its data-handling promises simultaneously, which is either an act of confidence or an act of ambition, and probably both.
What changes structurally is the definition of the front line in the AI assistant market. For years, the free tier was a funnel, a deliberately constrained experience meant to convert users into subscribers. The free tier no longer feels like a brief demonstration; it feels like a complete text assistant with premium doors surrounding it. The unlimited text offer applies to text chats; limits remain for file uploads, images, and other tools, and OpenAI notes the offering is subject to abuse guardrails. Deep Research, Agent Mode, and the upgraded GPT-5.6 Sol stay paid. The paid tier still has a compelling proposition, but OpenAI has now made the free tier good enough that staying free is a rational long-term choice for a very large portion of that billion-person base. That is a concession to gravity: at sufficient scale, the audience itself becomes the product, and metering it becomes a liability rather than an asset.
When a company with a billion weekly users decides the cost of keeping them is cheaper than the cost of losing them to a rival, the subscription model stops being a revenue strategy and starts being a loyalty tax for people who want fewer ads and a better model.