Some ChatGPT users have observed both performance improvements and response delays, fuelling speculation that OpenAI is privately testing GPT-5.6.

On June 19, blockchain outlet Decrypt reported that X was seeing a stream of test results from some paid subscription accounts that selected ChatGPT's GPT-5.5 Pro, saying it seemed as if a different model was effectively running.

Users commonly pointed to better performance and longer processing times. Some users said ChatGPT produced better results than before in web design and 3D game production. By contrast, tasks that the existing GPT-5.5 Pro typically finished in about 10 minutes took more than 1 hour, and other cases were reported to take 20 to 40 minutes.

Developer Ansh Chimal released a video comparing one-shot landing page outputs and claimed he was one of the lucky users granted early access to GPT-5.6 Pro. Dobroslav Radosavljevic, using Codex, said the model currently running felt completely different from the existing GPT-5.5 model.

Similar assessments emerged from 3D game generation tests. Connor Dart created a browser game with physics calculations and camera controls with a single prompt, and said it took 60 minutes 15 seconds to generate. He said, "It is not perfect, but it is very impressive as a result of testing game development with a single prompt." AI industry influencer Chetas Lua also wrote that he confirmed results that looked like a new model in robot simulation and 3D tests, adding, "Understanding has increased a lot compared with before, but work time started taking 20 to 40 minutes again."

It did not lead on every comparison. Chris, known as an AI benchmark account, said he compared the two models by entering a spaceship-building prompt, and the model believed to be GPT-5.6 Pro worked for 87 minutes, while GPT-5.5 Extra High took 34 minutes 42 seconds. He said, "GPT-5.6 may be incremental versus GPT-5.5, but it could amount to only simple improvement work," and added, "In some benchmarks it may be able to compete similarly with rival models, but it may not clearly pull ahead."

Leaked information also followed. An account believed to be leaker Pankaj Kumar said the knowledge cutoff date was pushed back to December 2025 and that an inference setting some testers call the "Juice Value" rose to 960 from 768. Another claim said its SVG and 3D design generation could surpass Anthropic's "Fable5" on certain tasks.

So far, OpenAI has not confirmed anything directly. The Information reported that Jakub Pachocki, OpenAI's chief scientist, was said to have mentioned internally that the next model was a meaningful improvement over GPT-5.5. That is read as a sign new model development is under way, but it does not confirm internal A/B testing within ChatGPT, a release timing, or specific specifications.

These observations are also growing amid changes in the competitive landscape. China's open-source model GLM-5.2 closed to within 1 point of Claude Opus4.8 in FrontierSWE, which evaluates long-form open-ended engineering tasks, and in the same test it was reported to have outperformed GPT-5.5. At the same time, Anthropic's main models Mythos5 and Fable5 are off the market after U.S. control guidelines. With a gap in top-tier models, observations have emerged that GLM-5.2 and the not-yet-public GPT-5.6 could target that space.

Possible price competition is also being discussed. OpenAI is reportedly reviewing token price cuts for developers and corporate clients, and there is also speculation that Anthropic could make a similar adjustment. Interest is also growing because both companies are in a phase of preparing for initial public offerings.

Whether it will be released remains uncertain. But on X, claims continue that when users select GPT-5.5 Pro in ChatGPT, a new model is being quietly tested on some accounts, and some speculate the public timing could be June 25. On prediction market Polymarket, bets piled into the possibility that GPT-5.6 will be released between June 22 and June 28, pushing the probability for the related contract as high as 89 percent at one point this week.

The core of the controversy is less about expectations of improved performance than that users say they first detected the possibility of a model swap inside the live service. Attention has turned to what mix OpenAI will offer between response speed and quality, and how quickly it will fill the gap left by rival models.

GPT 5.6 Pro continues to mogs Fable in 3d test working on games one shot too , for comment check quote post and also share your prompts and fable output i will tag you with 5.6 pro results credit : @mirochill https://t.co/kSMkWpOnyZ pic.twitter.com/6lsp3pUC6M