OpenAI has begun rolling out significant updates to its ChatGPT models, introducing new versions for both paid and free users, with a focus on improved accuracy, consistency, and user control. The updates, which started appearing for some users in early August 2026, aim to make the AI more direct and reliable across various tasks.
Paid subscribers to ChatGPT Plus and Pro are receiving GPT-5.6 Sol, a model specifically tuned for conversational use within the standard ChatGPT experience. OpenAI states that this version delivers more focused answers, adjusts its level of detail appropriately, and avoids unnecessary formatting. It is also designed to offer helpful corrections rather than simply agreeing with potentially flawed user prompts. For instance, simple questions are met with direct, single-line answers, while more complex inquiries receive detailed responses with relevant background.
A key enhancement for GPT-5.6 Sol users is a new "intelligence slider" that allows control over the model's reasoning effort. Users can choose between "Instant" responses, which are nearly immediate, and "High" reasoning, which may take several minutes but is intended for more complex tasks like research, coding, writing, planning, and decision-making.
OpenAI claims that GPT-5.6 Sol significantly reduces factual errors. Internal tests indicated a 68% reduction in responses containing at least one factual error compared to GPT-5.5 Instant. This improvement is particularly noted for handling dates, numbers, sources, legal rules, medical information, financial questions, and assumptions.
Free and Go subscribers are being upgraded to GPT-5.6 Luna, which will become the default model. Some users with Go subscriptions, priced around $10, have already observed this change. Starting the week of August 11, 2026, OpenAI plans to remove rate limits for text chats on these free and Go plans, though abuse protections will remain in place. Limits will still apply to other features such as file uploads and image generation.
Free users will also gain access to a new "Think" button. This feature allows GPT-5.6 Luna additional time to process and reason through more challenging questions, similar to the reasoning slider for paid users, but as a distinct user-initiated option. OpenAI's internal evaluations showed that GPT-5.6 Luna reduced responses with at least one factual error by 62% compared to GPT-5.5 Instant.
These updated models, GPT-5.6 Sol and GPT-5.6 Luna, are specifically designed for everyday conversational use within ChatGPT's regular chat interface. They do not affect the versions of GPT-5.6 Sol used in agentic workflows like Codex or ChatGPT Work.
In conjunction with these model upgrades, OpenAI is also implementing additional safeguards for users believed to be under 18. These protections include stricter boundaries regarding romantic roleplay, sexual content, dangerous activities, eating disorders, body image risks, age-restricted goods, and graphic violence. The rollout of these changes is gradual across the user base.
