Free users got the unlimited model. Paying users kept the one that is right more often.
The unlimited is real and text-only, behind abuse guardrails OpenAI does not enumerate. Files, images, and voice keep their caps. The 62% accuracy jump is OpenAI testing OpenAI, and the model tuned to be 'more reliable with facts' stays on the paid tier.
"OpenAI's ChatGPT update gives free users unlimited text chats on the new default model GPT-5.6 Luna, with factual errors 62% less common than the model it replaces; the rollout to Free and Go users completed in the week of August 10." [SOURCE ↗]
THE CLAIM. free ChatGPT users now get unlimited text chats on GPT-5.6 Luna, the new Free and Go default, with factual errors 62% less common than GPT-5.5-Instant, per OpenAI's August 6 announcement, fully rolled out since the week of August 10.
THE CHECK. the giveaway is genuine and the word unlimited arrives pre-qualified. OpenAI's own sentence caps it: unlimited text chats come 'subject to abuse guardrails', which are not enumerated anywhere, and 'Limits will still apply for file uploads, images and other tools.' The 62% is a relative figure from OpenAI's internal testing, not an independent evaluation, with no published sample or task mix. And the tier split does the quiet work: Luna is the fast, cheap variant; the model OpenAI describes as more reliable with facts, GPT-5.6 Sol, is the one Plus and Pro subscribers pay for. Free users got more quantity. The accuracy upgrade is the product being sold.
OpenAI announced the update on August 6 and staged it in three waves: the paid-tier GPT-5.6 Sol refresh landed the same day, GPT-5.6 Luna became the default for Free and Go users that week, and the headline benefit followed the week of August 10. The headline benefit, in OpenAI's own sentence: free
🔒 THE FULL AUTOPSY · FREE WITH AN ACCOUNTYou just read the free check. Sign in free, a code by email, no passwords, and the rest unlocks: the evidence trail, the steelman and the rebuttal, all 7 sources with quotes and screenshots, and our on-record call.
Couldn't verify your access — this looks like our error, not yours.