GPT-5.6 Sol gets an effort slider, and a 68% drop in responses containing a factual error
Plus and Pro users get a control over how much effort the model spends per response, replacing GPT-5.5 Instant. On an internal evaluation of financial, medical and legal prompts requiring factual detail, responses containing at least one factual error were about 68 percent less common with Sol and 62 percent less common with Luna than with GPT-5.5 Instant.
The slider is the more interesting shipment. Exposing compute-per-response as a user control concedes that there is no single correct point on the quality-versus-latency curve, and hands the trade to the person who knows what the answer is for.
On the accuracy number, read the construction carefully. It is an internal evaluation, on prompts selected for requiring factual detail, measuring the share of responses containing at least one factual error against a named predecessor. That is a well-specified claim and a meaningful improvement. It is not a claim about accuracy in general, and the domains chosen — financial, medical, legal — are the ones where a residual error rate still matters most.
Free users move to Luna as the default with unlimited text chats, which is the distribution half of the same announcement. The pricing move a week earlier makes the strategy legible: cheaper, more accurate, more widely available, in that order.
OpenAI — Improving GPT-5.6 Sol in ChatGPT and expanding access to GPT-5.6 Luna for free users → · OpenAI — GPT-5.6 August updates, deployment safety hub → · OpenAI — GPT-5.6: frontier intelligence that scales with your ambition →