OpenAI is removing limits on text-based chats for all ChatGPT users, eliminating one of the longest-standing distinctions between the free and paid tiers of its consumer product. The change was announced on August 6, 2026, alongside the rollout of the GPT-5.6 Luna model as the new default for ChatGPT Free and Go users, replacing GPT-5.5, and an upgraded GPT-5.6 Sol for Plus and Pro subscribers (TechCrunch, OpenAI).

The timing aligns with a milestone OpenAI recently crossed: ChatGPT surpassed 1 billion weekly active users (TechCrunch). The removal of text-chat caps applies across all tiers, but separate limits remain in place for files, images, voice, and image generation (TechCrunch).

For Free and Go users, the GPT-5.6 Luna model becomes the default. These tiers also gain a new "Think" button that lets users select higher reasoning power for complex questions — essentially telling the model to spend more computation before answering, which improves answer quality at the cost of a longer wait. For Plus and Pro subscribers, OpenAI is shipping an upgraded GPT-5.6 Sol model, tuned for quick tasks including questions, web research, advice, planning, writing, and decision-making. This version of Sol is distinct from the one used in Codex and Work environments, which remains unchanged (TechCrunch, OpenAI).

Plus and Pro users are also getting a thinking slider that adjusts how much computational deliberation the model applies before answering. The Free-tier Think button and the paid-tier slider serve a similar purpose — giving users explicit control over the tradeoff between reasoning depth and response speed — but at different granularity levels (TechCrunch, OpenAI).

OpenAI's internal evaluations report factual-error reductions versus GPT-5.5-Instant: 62% fewer for GPT-5.6 Luna and 68% fewer for GPT-5.6 Sol (TechCrunch). These are OpenAI's own benchmark results, not third-party evaluations, and the specific methodology and test sets have not been independently disclosed.

The rollout is staggered. The upgraded GPT-5.6 Sol is available to Plus and Pro users on the day of announcement. Changes for Free and Go users, including the Luna default model, arrive during that week. Unlimited text chats and the Think button for Free and Go users follow the week after (TechCrunch).

Two additional items appeared on OpenAI's newsroom the same day. The company published a post titled "Improving GPT-5.6 Sol in ChatGPT—and expanding access to GPT-5.6 Luna for free users" on August 6, and listed a partnership post titled "OpenAI and APA partner to advance responsible AI" with the same date. An engineering post on "Continuous voice interaction with GPT Live" was dated August 3 (OpenAI Newsroom.

Historically, OpenAI's tiering strategy has used message caps as the primary lever separating free and paid access. As of a May 2024 support page, ChatGPT Plus users had message limits up to five times greater than free users, with Team and Enterprise tiers higher still (OpenAI). The free tier's capabilities, default model, and limits have always been subject to change over time (OpenAI Help).

The removal of text-chat caps changes that calculus. The free tier now retains limits only on multimodal features — files, images, voice, and image generation — while text conversations become effectively uncapped. This repositions the paid tiers' value proposition around model quality (Sol versus Luna), reasoning controls (the thinking slider versus the Think button), and multimodal throughput, rather than raw text message count.

The broader context here matters for developers and enterprises. The distinction between the two Sol variants is an important detail: the upgraded GPT-5.6 Sol shipping to Plus and Pro is a ChatGPT-specific version, optimized for consumer interaction patterns like quick tasks, advice, and planning. The Codex and Work variant, which presumably serves more structured code-generation and enterprise workflows, is deliberately untouched. Teams relying on Sol through API or Work interfaces will not see behavioral changes from this update, which should reduce regression risk in production pipelines even as the consumer-facing model shifts.

The error-reduction figures, while notable, follow a familiar pattern in model releases. Vendors report internal evaluation improvements that are difficult to contextualize without access to the test sets or comparison methodology. A 62% reduction in factual errors sounds substantial, and it may well be, but the baseline (GPT-5.5-Instant) and the error taxonomy are not publicly specified. Developers evaluating whether to migrate workloads to Luna or the new Sol should treat these numbers as directional rather than definitive until independent benchmarks appear.

For the free-tier user base, now at a billion weekly actives, the practical change is straightforward. Text conversations no longer hit a cap. A Think button provides on-demand deeper reasoning. The default model is Luna, which OpenAI positions as more factually reliable than its predecessor. Whether that reliability holds under the volume and diversity of queries a billion users generate is a question the rollout itself will answer.