OpenAI is upgrading its free ChatGPT tier with a new model that cuts factual mistakes by 62 percent and removes the text message cap entirely, according to a report published this week by AI Weekly. The company is rolling out GPT-5.6 Luna as the default for free users starting this week, with unlimited text conversations and a new "Think" button arriving the week of August 10. The shift represents a strategic bet that winning the AI assistant market means making the free tier genuinely capable rather than treating it as a teaser for paid subscriptions.
According to the report, GPT-5.6 Luna produces factual errors 62 percent less often than GPT-5.5 Instant, based on OpenAI's internal testing across financial, medical, and legal prompts that require precision. A parallel paid-tier model called GPT-5.6 Sol, aimed at everyday tasks like quick questions, web research, planning, writing, advice, and decisions, shows a 68 percent reduction in factual errors compared to the same baseline. Free and Go users will get Luna as their default this week, with the unlimited text chat quota and Think button following next week. Paid Plus and Pro subscribers are being moved to Sol and receive a "thinking slider" that adjusts how much computational reasoning the model applies to any given question, while free users get the Think button on a per-message basis for complex queries.
The report frames the Think button and thinking slider as the new paid-tier justification once the free-tier message limit disappears. OpenAI is trading inference costs for retention and daily active user growth, the report states, positioning ChatGPT to win by becoming the default consumer interface rather than by pushing free users toward subscriptions. The accuracy figures come from OpenAI's own internal evaluations, not third-party benchmarks, so the report advises treating them as directional claims until independent testing confirms them. The reporting doesn't clarify where new caps will land on files, images, voice, and image generation, all of which remain metered, or how the unlimited text quota performs if usage spikes.
If Luna's accuracy gains hold up in real-world use, free users and cost-sensitive markets are the immediate winners, gaining access to a materially sharper model without paying, according to the report. That puts competitive pressure on Gemini and Meta AI to match the offer, since uncapping text chats on a tier most rivals still meter signals OpenAI believes ChatGPT wins through ubiquity rather than paywalls. The update positions the free tier as the front line of the assistant war, not just a demo, with the continuous reasoning slider and Sol model preserving a clear capability boundary between free and paid users. The real test will be whether independent benchmarks validate the 62 percent error reduction and whether the infrastructure can sustain unlimited free usage without introducing new friction points. The capability gap OpenAI is trying to hold—unlimited basic access versus continuous reasoning control—may define whether consumer AI becomes a product people subscribe to or a utility they expect for free, and whether competitors can afford to follow without burning through capital faster than they build retention.