I logged in to X early this morning and saw OpenAI's official account posting a tweet
There is no press conference hype or lengthy technical blog, but the information density is extremely high. Simply put, GPT-5.6 has been officially split into Sol and Luna versions, serving both paid and free users respectively, with substantial upgrades to the core experience.
Paid user: GPT-5.6 Sol launched, with a 68% drop in factual error rate
Starting today, the chat interface for Plus and Pro users has automatically switched to GPT-5.6 Sol.
The core indicator of this upgrade is not score running, but factual accuracy. OpenAI provided a very specific test data: in evaluations of high-risk fields such as finance, medicine, and law, GPT-5.6 Sol showed a 68% reduction in the number of factual error responses compared to GPT-5.5 Instant.
For users who view ChatGPT as a productivity tool rather than just a chat toy, this number is more convincing than any "smarter" description. It means that the cost of cross validation can be significantly reduced when dealing with serious tasks.
Another practical update is the inference effort slider. Plus/Pro users can now manually adjust the depth of reasoning for each reply - simple questions are answered in seconds, and complex questions fill their thinking budget. This essentially transforms the 'thinking mode' from a switch to a continuous variable, making the transition between daily conversations and professional analysis smoother.
Note the boundary: The GPT-5.6 Sol updated this time is only available for Chat chat scenarios. The model versions behind Work and Codex remain unchanged, don't expect the same improvements in the workflow.
Free/GO users: Luna unlimited chat+Think button placement
This may be the most user-friendly aspect of this update for ordinary users.
Starting tomorrow, Free and Go users will have unlimited text chat with GPT-5.6 Luna. It's no longer a daily quota system, but a truly infinite dialogue.
More importantly, the 'Think' button is open to free users. When encountering difficult problems that require multi-step reasoning, you can actively trigger deep thinking mode, allowing the model to invest more computing power to "think clearly before answering". Previously, this was exclusive to the o1/o3 series or paying users, but now it has become a part of the basic experience.
The intelligence of Luna itself has also been upgraded. Although the official has not disclosed a specific benchmark, the combination of "unlimited chat+triggered thinking" has significantly raised the practical ceiling of the free version.
The product logic behind this update
Reading through the whole tweet, we can clearly perceive the current strategic direction of OpenAI: instead of pursuing the ultimate performance of a single model, we will experience inclusive benefits through model layering.
Sol is aimed at paying users: emphasizing accuracy, controllability, and reliability in professional scenarios;
Luna is aimed at the general public: emphasizing accessibility, basic intelligence, and democratization of thinking ability.
The two share the technical foundation of GPT-5.6 generation, but have made differentiated optimizations for completely different usage expectations. This distribution strategy of "same generation but different entities" not only avoids excessive consumption of computing power by free users, but also ensures the trust of paying users in high-risk scenarios.
As for whether the actual tactile sensation matches the official data, it is recommended that paying users try the response differences of the slider in different gears today, and free users can test the effect of the Think button tomorrow by taking a few hard questions that they couldn't answer before
Reminder: This article is based on OpenAI's official X platform tweets and does not constitute any investment or usage advice.
