Three ChatGPT leaks in five hours were one release note — and the line none of them quoted is the real change

Three separate leaks landed in five hours claiming three separate ChatGPT changes: an upgraded GPT-5.6 Sol for Plus and Pro, a new reasoning slider rolling out, and unlimited chats coming for free users. All three are real. They are also all one release, and OpenAI published it themselves — so instead of grading the leaks we can just read the primary document.

What OpenAI's own post says, dated today.

Plus and Pro get an updated GPT-5.6 Sol in the chat experience, available starting today, with a slider on web, mobile and desktop that sets how much thought goes into each answer. Free and Go users get GPT-5.6 Luna as their default model this week; starting next week they also get unlimited text chats and a Think button, subject to abuse guardrails, with limits still applying to file uploads, images and other tools.

The single most substantive line was in none of the leaks: "the same model now powers both Instant responses and deeper reasoning" for Plus and Pro. That is the actual change. ChatGPT's Instant and Thinking modes have behaved like two different models with two different personalities, and moving to higher effort meant switching products mid-conversation. One model across the whole range is an architecture decision, not a feature — and it is what the slider is for. A slider between two different models would just be a model picker with a nicer coat of paint.

The 68% figure is real, and its baseline deserves reading twice. OpenAI reports that in an internal evaluation of financial, medical and legal prompts requiring factual detail, responses containing at least one factual error were about 68% less common with GPT-5.6 Sol — and about 62% less common with GPT-5.6 Luna — than with GPT-5.5 Instant. Two things follow. The comparison is against 5.5 Instant, a cheaper model from the previous generation, not against the GPT-5.6 Sol people are using right now; there is no published delta against the model this replaces. And Luna's 62% went unmentioned everywhere, which matters because Luna is the model free users are about to get unlimited access to. It is an internal eval, so nobody outside has reproduced either number.

"Unlimited" is unlimited Luna. Free and Go users are getting uncapped text chat on OpenAI's cheapest tier — $0.10 per million input, $0.60 output on the API, the model that took an 80% price cut on July 30. One outlet framed the same release as restricting free users to the weakest model, and both readings are fair from the same paragraph: the default becomes Luna, and the cap comes off. Whether that is expanded access or a cheaper floor depends on what free users were reaching before, which OpenAI does not say.

For anyone building on this: nothing changed. OpenAI is explicit that this version of Sol is only in the Chat experience, and that the Sol powering Work and Codex is not changing. We read the OpenRouter catalog to check: openai/gpt-5.6-sol and gpt-5.6-luna are still carrying their original listing date of July 9 with unchanged prices — $5/$30 and $0.10/$0.60 per million. No new id, no new snapshot, nothing to migrate. If you evaluate models by API behaviour, today was a no-op; if you use ChatGPT, today was the bigger of the two.

One thing worth watching rather than reporting: the post says the update makes the model offer "a helpful correction when simply agreeing wouldn't be useful". Reduced sycophancy is easy to claim and hard to measure, and unlike the 68% it comes with no number at all.