OpenAI Updates ChatGPT GPT-5.6 Sol to Enhance Fact Reliability
OpenAI announced an update to GPT-5.6 Sol for Plus and Pro users, making responses more focused, improving fact reliability, and reducing unnecessary formatting and details.
The same model supports both instant responses and deep reasoning, with users able to adjust the level of thought using a new slider. Internal evaluations show that for financial, medical, and legal questions, the rate of responses with at least one factual error decreased by about 62% for GPT-5.6 Luna compared to GPT-5.5 Instant, and by about 68% for GPT-5.6 Sol. GPT-5.6 Luna became the default model for Free and Go users this week, and starting next week, free users will have unlimited text chat and can use the Think button for deep questions, though still subject to anti-abuse restrictions.
This is a routine optimization event for the model's user experience, reallocating computing power and reasoning resources towards higher reliability and controllable depth. Plus and Pro users benefit from enhanced focus and factual accuracy, while free users gain broader text access and reduced risk of errors in professional queries.
Source: Public Information
ABAB AI Insight
Since the launch of the GPT-5.6 series, OpenAI has been continuously fine-tuning everyday conversation layers, positioning Sol as the flagship experience that unifies instant and deep reasoning. Previous models' discrepancies in factual accuracy have been quantified by internal evaluations, showing that targeted training significantly reduces hallucinations in professional fields.
On the capital path, through layered models (Sol/Luna) and user-controllable sliders, high-cost deep reasoning is allocated on demand rather than fixed high reasoning. The motivation is to expand the free user base while capturing high-value interactions with the paid tier and reducing the trust costs associated with factual errors.
Similar to the previous parallel iteration path of the o series reasoning models and standard models, it is currently transitioning from capability release to a balance of reliability and controllability.
Essentially, this is a technological replacement: fixed deep reasoning is replaced by user-adjustable thinking intensity, mechanism-wise shifting the computational budget from a default high cost to on-demand allocation to balance scale and quality.
ABAB News · Cognitive Laws
The decrease in factual error rates is more valuable than parameter scale.
Controllable depth retains users better than fixed depth.
Unlimited chat in the free tier is the best advertisement for the paid tier.