In short
- X users are showing sharp output and long response times in ChatGPT and Codex this week.
- Some users have reported improved web design and 3D game output from ChatGPT.
- The official release of GPT-5.6 is rumored for next week, but OpenAI has not announced any plans.
Something felt different in ChatGPT this week—and a lot of people noticed it at once.
Across X, testers have spent the last two days exchanging screenshots and timers, all pointing to one idea: OpenAI is quietly A/B testing GPT-5.6 within ChatGPT, modified by some users who choose GPT-5.5 Pro.
Developer Anshu Chimala posted a side by side video On Thursday comparing pages that were shot once, he writes: “Well, I’m one of the lucky ones with the initial access to GPT-5.6 Pro.”
Producer Dobroslav Radosavljevič written on X that whatever he’s running inside Codex, the OpenAI agent, “feels waaaaaaaaay different from (a) 5.5 model.” Answers are divided between believers and people who call them placebo.
A clear example of all images is time. Conor Dart, one of the many X users developing the rumor, ran the test with a 3D browser game with physics and camera tools – which only took an hour to create, against the GPT-5.5 Pro’s usual 10 minutes.
“It’s not perfect, but for one game AI test, this is pretty cool,” wrote Dart.
AI inside Chetas Lua report a little test of the same simulation, and I am sure that the results are from the new version of OpenAI: “GPT 5.6 Pro continues to mog (The Anthropic Myth 5) in the 3D test,” he he wrote. “Working again on one game.”
In another post, he noted response times of up to 20 or 40 minutes, speeds he said hadn’t been seen since before GPT-5.5 was deployed.
Not every comparison attracted the rumor model. Chris, the benchmark representative of AI on X, gave the two models a similar process for creating ships – the supposed GPT-5.6 Pro worked for 87 minutes against the GPT-5.5 Extra High’s 34 minutes and 42 seconds.
“As I said before, based on the main authority, GPT-5.6 will be an improvement/strength on GPT-5.5, not a killer of Fable,” he said. he wroteI realize Fable 5 still beats both models on space geometry. “My rough expectation was that it would compete with Fable 5 on some benchmarks, maybe half as well depending on the group, but not all of them.”
Another post made by leaker Pankaj Kumar detailed leak Further: the reduction of information beyond December 2025, a test-test that some testers call “The need for Juice” is said to have risen to 960 from 768, with SVG and 3D creating a generation powerful enough to beat Fable 5 in some tasks.
Nothing comes from OpenAI—but the details are consistent in all accounts: strong thinking, unfinished business, and a release called Kindle-Alpha.
The AI assistant known as Leo, citing unnamed sources, wrote in the thread that the supposed version “is now being tested in a fake way where 5.5 Pro is selected in ChatGPT,” especially for some Pro accounts, and will launch publicly next Thursday, June 25.
The closest OpenAI gets to fingering any of this is a memo, not a tweet. Chief scientist Jakub Pachocki is said to have told staff that the next generation is meaningful change on GPT-5.5, according to the source report The Information. It’s not a guarantee of A/B testing, a release date, or anything else floating around X, but it does confirm that a new version is in development.
Decrypt reached out to OpenAI to ask if GPT-5.6 is being tested within ChatGPT, but the company did not respond by press time.
Why OpenAI can be fast
If OpenAI is rushing a new model out the door, it has reasons. The Chinese version of the opening GLM-5.2 it follows the Claude Opus 4.8 by just one point on FrontierSWE—a benchmark that rates AI assistants for multi-hour, open-source engineering projects based on the amount of power—beating the GPT-5.5 on the same test.
Anthropic, meanwhile, is dealing with the damage it causes. The company’s most popular models Mythos 5 and Fable 5 remain under the US export control instructions which was released on June 12 for jailbreak threats, leaving a gap in the market that GLM-5.2 and the hypothetical GPT-5.6 both have a chance to fill.
If Anthropic CEO Dario Amodei and President Donald Trump make peace, then Fable 5 will be more powerful than any other model available, and the positive difference between the top version of Anthropic and OpenAI is growing more than ever.
There is also money on the table. OpenAI says measuring prices on those indicators money development and businesses, according to Wall Street Journalexpecting Anthropic to do the same as all companies preparing for dueling IPOs.
Whether or not anything adds up to the actual release of GPT-5.6 is something only OpenAI can confirm, and the company has remained silent for a week of leaks and reports of secret testing. Polimamarket traders are not waiting, however – contracts to settle between June 22 and June 28 rose to 89% this week.
Daily Debrief A letter
Start each day with top stories right here, including originals, podcasts, videos and more.
