OpenAI's GPT-5.6 Luna cuts errors by 62% as free ChatGPT gets unlimited text

The latest ChatGPT rollout signals a pivotal shift in AI accessibility: by unshackling free text usage and baking advanced reasoning into the default model, OpenAI is challenging the industry’s free-vs-premium divide. For AI practitioners, the introduction of a ‘thinking slider’ and granular error reduction metrics provides a rare window into how model tuning translates to real-world reliability.

OpenAI is making a dramatic move to expand the reach and reliability of ChatGPT by removing text-based chat limits for free users while simultaneously rolling out more capable, less error-prone AI models across all tiers. Announced on August 6, 2026, the changes center on the new GPT-5.6 model family – Luna for Free and Go users, and Sol for Plus and Pro subscribers – which achieve substantial reductions in factual errors compared to the prior GPT-5.5-Instant. The company is also introducing a “Think” button for free users to tap higher reasoning on demand, and a thinking slider for paid tiers that lets users dial in the depth of the model’s processing. These updates come as ChatGPT has crossed 1 billion weekly users, underscoring the platform’s gravitational pull and OpenAI’s need to serve an increasingly diverse user base without compromising quality.

“

This move leverages the new GPT-5.6 Luna model, which is 62% less likely to produce factual errors than its predecessor.

The shift to unlimited free text chats is not merely a pricing or access change; it fundamentally alters the competitive dynamics in the AI assistant market. Until now, heavy users on free tiers could hit rate limits, nudging them toward subscriptions or alternatives. By uncapping text-based interactions, OpenAI is effectively making ChatGPT the default, always-available text companion for hundreds of millions of users worldwide. This move leverages the new GPT-5.6 Luna model, which is 62% less likely to produce factual errors than its predecessor. Improved reliability is critical for retaining trust, especially when the assistant is used for research, learning, and decision-making. The “Think” button adds a layer of optional sophistication, allowing free users to selectively invest additional compute for complex problems without paying a subscription fee, a design that balances cost management for OpenAI while empowering users.

For paid tiers, the upgraded GPT-5.6 Sol model further pushes the envelope. Internal evaluations show a 68% reduction in factual mistakes, and the model delivers more direct, concise answers. The thinking slider gives Plus and Pro users fine-grained control over the reasoning effort, from quick-fire factual lookups to multi-step problem solving. This granularity is a step toward more transparent and customizable AI interactions, potentially reducing hallucination by letting users signal the required depth of analysis. The slider also hints at a broader industry trend toward exposing model “cognition” settings, moving from opaque black boxes to adjustable intelligent agents.