GPT-5.6 update brings Luna to free users and an effort slider
OpenAI says the updated GPT-5.6 Sol makes roughly 60% fewer factual errors. Free and Go users get a new default model.
Who gets what
In the August 6 update, Plus and Pro users got an updated GPT-5.6 Sol with a slider to choose how much effort ChatGPT spends on a response. Free and Go users got GPT-5.6 Luna as the new default, with a Think button for harder questions.
Measured change
OpenAI reports roughly 60% fewer factual errors for Sol across three prompt sets, plus gains of 15.6 points on HealthBench Professional and 8.5 on HealthBench Hard. It also published dedicated under-18 safety evaluations for the first time.
More articles
News · Models