Something felt off about ChatGPT this week — and a ton of people noticed at the same time.
Across X, testers spent the last couple days swapping screenshots and stopwatch timings, all pointing at one theory: OpenAI is quietly A/B testing GPT-5.6 inside ChatGPT, swapping it in for some users on GPT-5.5 Pro.
Developer Anshu Chimala posted a side-by-side video on Thursday comparing one-shot landing pages. His caption: “Well well well, I’m one of the lucky ones with early GPT-5.6 Pro access.”
One-shot pages from 5.5 Pro vs 5.6 Pro. Can you tell which is which? Looks like OpenAI’s finally getting somewhere with design.
Not everyone’s convinced it’s real. Replies underneath these posts split pretty evenly between believers and people calling it placebo. But the numbers are hard to ignore.
Conor Dart ran a one-prompt 3D browser game — with physics and camera controls — that took just over an hour to generate. GPT-5.5 Pro usually knocks that out in 10 minutes. “Not perfect, but for a one-prompt AI game dev test, this is seriously impressive,” Dart wrote.
AI insider Chetas Lua reported response times stretching to 20 or 40 minutes, a pace he said hadn’t shown up since before GPT-5.5 shipped. “GPT 5.6 Pro continues to mog Fable 5 in 3D test,” he wrote.
Not every comparison flattered the rumored model though. One AI benchmarker gave both models the same spaceship-building prompt — the suspected GPT-5.6 Pro worked for 87 minutes against GPT-5.5 Extra High’s 34 minutes and change. Fable 5 still beat both on the spaceship’s core geometry.
Leaker Pankaj Kumar detailed specs: a knowledge cutoff pushed to December 2025, a reasoning-effort setting raised to 960 from 768, and SVG and 3D design generation that allegedly beats Fable 5 on some tasks.
AI influencer Leo, citing unnamed sources, wrote that the model is “now being stealth tested when 5.5 Pro is selected in ChatGPT” for some Pro accounts, with a planned public launch on Thursday, June 25.
The closest thing to an official OpenAI fingerprint is a memo. Chief scientist Jakub Pachocki reportedly told staff the next model is a meaningful improvement over GPT-5.5, according to The Information. That’s still not confirmation of A/B testing or a release date, but it does confirm something’s been brewing.
Decrypt reached out to OpenAI for comment. The company didn’t respond by publication time.
So why the rush? China’s open-source GLM-5.2 trails Claude Opus 4.8 by just one point on FrontierSWE while beating GPT-5.5 outright. Anthropic’s flagship Mythos 5 and Fable 5 models remain pulled under a U.S. export control directive issued June 12. That’s a gap at the top of the market that GLM-5.2 and a hypothetical GPT-5.6 are both positioned to fill.
OpenAI is also reportedly weighing price cuts on the tokens it charges developers and enterprises. The competitive pressure is real, and the window might not stay open long.
