OpenAI has refreshed the default models that power ChatGPT, introducing the GPT-5.6 series across free and paid tiers. The update aims to refine response quality, give users more control over reasoning depth, and expand access for those on the free plan.
For ChatGPT Plus and Pro subscribers, GPT-5.6 Sol becomes the new default in both instant and reasoning modes. OpenAI describes the model as tuned to produce more focused answers that adjust their level of detail according to the question and reduce unnecessary formatting. The company also claims improved factual reliability. In internal tests involving financial, medical and legal prompts that required precise details, responses containing at least one factual error were 68 percent less common than with the previous default, GPT-5.5 Instant. A new slider has been added on web, mobile and desktop so paid users can decide how much reasoning the model applies to a given reply.
Free and Go users receive GPT-5.6 Luna as their new default. Alongside the model change, these accounts gain unlimited text chats. A Think button for more complex queries is also arriving. Luna is scheduled to reach free users this week, with the unlimited text allowance and Think button following next week. Existing limits on image generation, file uploads and similar features remain in place.
The Sol variant rolling out for paid users is optimized specifically for the Chat experience and will not alter the versions used in ChatGPT Work or Codex. The changes begin arriving immediately for Plus and Pro accounts, while free-tier updates follow a staggered schedule.
These adjustments continue OpenAI’s pattern of frequent default-model refreshes. The previous shift from GPT-5.3 Instant to GPT-5.5 Instant occurred only in May. By pushing a more capable model and removing text-chat caps for free users, the company is attempting to keep casual users engaged while differentiating the paid experience through finer control and claimed accuracy gains. Internal accuracy figures are difficult to verify independently, and “unlimited” text still sits alongside other usage restrictions. The rapid succession of model names and feature tweaks also risks leaving users uncertain about what exactly has changed from one week to the next.
In a competitive landscape where several frontier models now exist, the default that serves more than a billion weekly users carries particular weight. OpenAI’s latest move prioritizes response focus, selective reasoning depth for paying customers, and broader access for everyone else. Whether the accuracy improvements hold up outside controlled tests, and whether unlimited text meaningfully changes free-tier behavior, will become clearer as the models reach full availability.
