AI-generated summary: OpenAI has rolled out a major update to ChatGPT, switching the default model for free users to GPT-5.6Luna, with no longer any text chat limits. Plus and Pro users receive the optimized GPT-5.6Sol version, featuring significantly improved factual accuracy and response quality; internal evaluations show a 68% reduction in hallucination rates compared to the previous generation. The new model dynamically adjusts response depth based on question complexity, providing direct answers for everyday queries and comprehensive responses for complex tasks. Additionally, a thought volume slider has been introduced to ensure a more consistent and seamless chat experience for one billion weekly users. Zhang Yiming recently stated at a Seed all-hands meeting that programming is merely one of today’s hot trends and should not dominate one’s focus—greater opportunities lie beyond programming.Source: AIBase
OpenAI today rolled out a major update to ChatGPT.
Free users will have their default model switched to GPT-5.6Luna, with no limits on text conversations in Chat mode, allowing unlimited use. This feature will be gradually rolled out to all free users this week.
Plus and Pro users receive the same upgrade. GPT-5.6Sol in Chat has been specially optimized, with the official team stating significant improvements in factual accuracy and response quality. Altman confirmed on X: "5.6Sol performs much better in chat—free users can now enjoy unlimited text chat!"
OpenAI states that one billion people per week use ChatGPT for quick Q&A, web searches, planning, research, consultation, and complex decision-making. This GPT-5.6Sol update is designed to cover the entire spectrum of these use cases. The updated model provides more focused responses, adjusts its level of detail according to the question, avoids unnecessary formatting clutter, and offers helpful corrections when mere agreement would be unproductive. For Plus and Pro users, the same model powers both instant answers and deep reasoning, delivering a more consistent experience.
Specifically, when asked quick questions, the model provides direct answers with necessary context; for complex tasks such as multi-step planning, research, or writing, it delivers more comprehensive responses while keeping core recommendations clear. In OpenAI’s comparative examples, in a cycling consultation scenario, GPT-5.5Instant lists all details including rain, wind, and temperature, whereas GPT-5.6Sol first identifies the user’s true concern—recognizing that wind, not rain, is the primary factor—and retains only the details relevant to the cyclist, even generating a customized weather forecast.
There has also been significant improvement in hallucinations. Internal OpenAI evaluations show that, in fact-based prompts across finance, healthcare, and legal domains, GPT-5.6Luna exhibits at least one factual error in approximately 62% fewer cases than GPT-5.5Instant, while GPT-5.6Sol shows a reduction of approximately 68%.
Subscribers also receive an updated UI experience. Previously, after the launch of the three models in version 5.6, users had to select both a model and inference intensity simultaneously, making the process complex. Now, Chat has introduced a slider feature identical to Codex’s, allowing Plus and Pro users to adjust the model’s thinking intensity via slider on web, mobile, and desktop—keeping it in Quick mode for everyday questions and sliding higher when planning, researching, writing, programming, or making decisions.
Free users receive full access to GPT-5.6Luna with unlimited text chat, but features such as image generation and file uploads remain limited. Since Luna is not a general-purpose model, OpenAI has equipped it with a new "Think" button to handle complex queries.
It is worth noting that this update to GPT-5.6Sol is limited to the chat experience in ChatGPT; the versions of GPT-5.6Sol used in Work and Codex are not included in this update.
This update stands out against the industry’s widespread focus on coding skills. Recently, Zhang Yiming, founder of ByteDance, also stated at the Seed all-hands meeting that programming is merely one of today’s hot topics and should not completely dictate one’s direction—there are even greater opportunities beyond coding. OpenAI, having already established itself among the top tier in coding capabilities, continues to invest effort in optimizing its chat mode, likely due to its large consumer user base and advertising-driven business model.