Leaked screenshots and canary traces suggest OpenAI is testing GPT‑5.6

OpenAI has not confirmed GPT-5.6. That is the first fact to keep in view as screenshots and “canary” traces tied to a supposed next model circulate across blogs and social feeds in late May 2026. The chatter points to two things: an alleged cleaner ChatGPT interface and backend references that some observers say appeared in Codex-related logs. But the public evidence so far is indirect, and the company’s own product pages still stop at GPT-5.5. (openai.com) The official record matters here because OpenAI did make a real release one month ago. (openai.com) On April 23, 2026, the company introduced GPT-5.5 and said it was rolling out across ChatGPT and Codex, with API availability following on April 24. Its help documentation, updated two days ago, says GPT-5.5 Instant is now the default for logged-in ChatGPT users. ### So what, exactly, leaked? ChatForest said on May 24 that GPT-5.6 “hasn’t been announced” but had “surfaced in OpenAI’s Codex backend logs,” framing that as evidence of active testing. (openai.com) Startup Fortune and other blogs described screenshots that purportedly showed a cleaner interface and backend traces linked to Codex. None of those reports included an OpenAI confirmation. A separate Startup Fortune report from April 22 described an earlier, different episode involving a model labeled “gpt-5.5-turbo-preview” that was briefly visible in OpenAI’s Codex tool for about 90 minutes before disappearing. (openai.com) That report said screenshots spread on Reddit and X and that the endpoint later returned a “model not found” error. ### Why do people keep focusing on Codex traces? (chatforest.com) OpenAI itself tied GPT-5.5 closely to Codex in its April 23 release. The company said GPT-5.5 was rolling out in ChatGPT and Codex, and said the model showed gains in coding, computer use, knowledge work and tool use. It also said GPT-5.5 used fewer tokens on Codex tasks than GPT-5.4. That makes Codex a plausible place for pre-release sightings, at least as an inference. But it remains an inference. The public reporting around GPT-5.6 relies on screenshots, log mentions and rumor trackers rather than release notes, system cards or product documentation from OpenAI. (startupfortune.com) ### Does the “cleaner interface” claim fit anything OpenAI has already shipped? OpenAI’s own help page says GPT-5.5 outputs are “more streamlined,” with “cleaner formatting” and less unnecessary header text. (openai.com) That means at least part of the leak narrative lines up with changes OpenAI has already publicly described for the current model generation. That does not prove the screenshots are fake or real. It does mean a cleaner-looking interface, by itself, is weak evidence of a new numbered model. (chatforest.com) Interface polish can reflect ongoing product tweaks, staged experiments or routing changes inside an existing release. That is an inference based on OpenAI’s documented GPT-5.5 behavior and the absence of any GPT-5.6 announcement. ### What should boards, founders and investors take from this? May 24’s leak reports are best treated as a signal of possible internal testing, not as a product roadmap. (help.openai.com) OpenAI’s official pages currently document GPT-5.5, not GPT-5.6, and the company has not published a system card, pricing page or launch post for any 5.6 model. That distinction matters for companies building on OpenAI. Product plans, pricing assumptions and customer commitments are usually anchored to official availability, usage limits and documented model access — the sort of details OpenAI included for GPT-5.5 in its release and help pages. (openai.com) ### Where would real confirmation show up first? OpenAI’s April 23 product post and its ChatGPT help documentation show the pattern to watch: named model announcement, rollout language, tier availability and, for major releases, updated safeguards or system-card material. (openai.com) As of May 24, 2026, those sources still point to GPT-5.5 as the latest confirmed model family in ChatGPT. The next concrete checkpoint is therefore not another screenshot. It is whether OpenAI adds a GPT-5.6 release page, updates its help center model documentation, or publishes API availability and safety materials under that name. (openai.com)