This week, several users posted on platform X that the performance of ChatGPT and OpenAI programming tool Codex has changed: stronger answers and more complete results are generated, but waiting times are significantly longer. Around these phenomena, the outside world began to speculate that OpenAI was conducting small-scale tests of GPT-5.6 within ChatGPT, and that some users who selected GPT-5.5 Pro might have been switched to a new model.
Some developers posted the results of the comparison, stating that the new version performed better in web design, 3D game generation and code understanding. It was also mentioned that, under the same hint, the time required for the system to produce the complete result was about 10 minutes, stretching to 20 minutes, 40 minutes, or even more than an hour.
Design and 3D tasks are raised
The developer Anshu Chimala publishes a comparative video showing the landing page generated under the same hint, stating that you may have been using GPT-5.6 Pro in advance. Another developer, Dobroslav Radosavljevič, also stated that the currently operating model of Codex “did feel very different from 5.5”.
Conor Dart tests a task for a single hint to generate a 3D browser game that contains physical effects and camera control. He stated that the complete production process took 60 minutes and 15 seconds, which was clearly slower than the GPS-5/Pro response, which was usually about 10 minutes, but the result was still surprising to him.
Long-term response to common features
The most consistent signal at present is not a change in the interface, but a time-consuming rise. Several users stated that the suspect new model would be deduced over a long period of time in complex missions, re-emerging the slow response that was common in early high-level models.
However, not all comparisons support the judgement of “obvious upgrading”. An AI benchmark tester on the X platform stated that in the same ship ' s modelling mission, the version suspected to be GPT-5.6 Pro was in operation for 87 minutes, while GPT-5.5 Extra High was in use for 34 minutes 42 seconds. Anthropic Fable 5 is still better than both in terms of core geometry.
OpenAI has not confirmed release time
More statements were also made that the knowledge deadline could advance to December 2025, with some of the test parameters being adjusted upwards and SVG and 3D design capabilities being enhanced. This information was not confirmed by OpenAI.
An AI-based blogger quoted unnamed sources as saying that OpenAI might be in the process of placing the model behind cover for some Pro users, on public release time or on 25 June. It was also reported by The Information that the principal scientist in OpenAI, Jakub Pachocki, had suggested to his staff that the next generation model would be “meaningly higher” than GPT-5.5.
Competition rises and increases market interest
Market interest in GPT-5.6 is also linked to the recent rise in competition for large models. It was reported that the open-source model GLM-5.2 in China was close to Claude Opus 4.8 and exceeded GPT-5.5 in some of the engineering-type baseline tests.
At the same time, Mythos 5 and Fable 5 of Anthropic are still not reinstated by a United States export control directive of 12 June. This leaves a gap in the competition for the market head model and gives more attention to the next generation of OpenAI models.
Additional information:According to the Wall Street Journal, OpenAI is also considering downward token pricing for developers and business clients. The report also mentions that Polymucket on the contract issued between 22 and 28 June for GPT-5.6 was once priced to 89 per cent this week.