Users Report Noticeable Changes in ChatGPT, Hinting at GPT-5.6 Testing
A palpable shift in ChatGPT’s performance this week has not gone unnoticed by its user base. Across social media platforms, particularly X, users have been sharing screenshots and timing tests, coalescing around a prevailing theory: OpenAI is discreetly conducting A/B tests of a new model, GPT-5.6, within ChatGPT, offering it to select users who typically access GPT-5.5 Pro.
Developer Anshu Chimla posted a comparative video on Thursday showcasing one-shot landing pages, accompanied by the caption: “Well well well, I’m one of the lucky ones with early GPT-5.6 Pro access.”
Developers Observe Performance Shifts
Developer Dobroslav Radosavljevič also took to X, describing the experience within Codex, OpenAI’s coding agent, as feeling “waaaaaaaay different than [the] 5.5 model.” While some users in the replies expressed belief, others dismissed the observations as a placebo effect.
A consistent theme emerging from these discussions is the significant increase in processing time. Conor Dart, an X user who has been amplifying these rumors, conducted a test generating a one-prompt 3D browser game complete with physics and camera controls. The task, which took just over an hour to complete, starkly contrasted with GPT-5.5 Pro’s typical 10-minute generation time.
“Not perfect, but for a one-prompt AI game dev test, this is seriously impressive,” Dart remarked.
AI Insiders Weigh In on Potential New Model
AI insider Chetas Lua reported a similar slowdown while testing a robotic simulation, expressing strong conviction that the results stemmed from OpenAI’s new model. “GPT 5.6 Pro continues to mog [Anthropic’s Fable 5] in 3D test,” Lua wrote. “Working on games one shot too.” In a separate post, Lua noted response times extending to 20 or 40 minutes, a pace not seen since before the release of GPT-5.5 Pro.
However, not all comparisons favored the rumored model. Chris, an AI benchmarker on X, presented a spaceship-building prompt to both models. The suspected GPT-5.6 Pro took 87 minutes, compared to GPT-5.5 Extra High’s 34 minutes and 42 seconds.
“As I’ve said before, based on great authority, GPT-5.6 will be an incremental/solid improvement over GPT-5.5, not a Fable killer,” Chris stated, while also noting that Fable 5 still outperformed both models in the spaceship’s core geometry. “My rough expectation has been that it would trade blows with Fable 5 on some benchmarks, maybe win around half depending on the category, but not clearly surpass it overall.”
Leaked Details Suggest Advanced Capabilities
Further details emerged from a post attributed to leaker Pankaj Kumar, suggesting a knowledge cutoff extended to December 2025. The leaks also indicated a potential increase in the “Juice Value,” a reasoning-effort setting, from 768 to 960. Additionally, the generation of SVG and 3D designs is reportedly robust enough to surpass Fable 5 in certain tasks.
While these details have not been officially confirmed by OpenAI, they align with user observations of enhanced reasoning capabilities, an unfinished front-end, and a release candidate codenamed Kindle-Alpha.
Stealth Testing and Anticipated Launch
AI influencer Leo, citing unnamed sources, suggested in a thread that the suspected model is currently undergoing “stealth testing” for select Pro accounts when GPT-5.5 Pro is chosen in ChatGPT. A public launch is reportedly slated for the following Thursday, June 25.
OpenAI’s Internal Communications and Market Pressures
The closest indication of an official OpenAI stance comes from a memo reportedly sent by Chief Scientist Jakub Pachocki to staff, describing the next model as a “meaningful improvement” over GPT-5.5, according to a report by The Information. This memo, however, does not confirm A/B testing, a release date, or any specific technical details circulating on X, but it does validate the development of a new model.
Decrypt reached out to OpenAI for comment regarding the testing of GPT-5.6 within ChatGPT, but received no response by the time of publication.
Several factors may be driving OpenAI to potentially accelerate the release of a new flagship model. China’s open-source model GLM-5.2 is closely trailing Claude Opus 4.8 on the FrontierSWE benchmark, a metric assessing AI agents on multi-hour, open-ended engineering projects, and has already surpassed GPT-5.5 on the same test.
Meanwhile, Anthropic is navigating its own challenges. The company’s leading models, Mythos 5 and Fable 5, have been temporarily withdrawn following a U.S. export control directive issued on June 12, due to a disputed jailbreak vulnerability. This situation has created an opening in the high-end market that both GLM-5.2 and a potential GPT-5.6 are positioned to exploit.
Furthermore, financial considerations are at play. The Wall Street Journal reports that OpenAI is reportedly considering price reductions on the tokens it charges developers and enterprises, anticipating similar moves from Anthropic as both companies prepare for potential IPOs.
Whether these developments culminate in an official GPT-5.6 release remains to be seen, as OpenAI has maintained silence amidst a week of leaked information and stealth-testing claims. However, traders on Polymarket are actively speculating, with contracts for a launch between June 22 and June 28 reaching as high as 89% this week.
Leave a Reply