In short
Customers throughout X report sharper outputs and unusually lengthy response instances in ChatGPT and Codex this week.
Some customers reported considerably improved net design and 3D online game outputs from ChatGPT.
A proper launch for GPT-5.6 is rumored for subsequent week, however OpenAI has but to announce plans.
One thing felt totally different in ChatGPT this week—and lots of people seen directly.
Throughout X, testers spent the previous two days swapping screenshots and stopwatch instances, all pointing to at least one concept: OpenAI is quietly A/B testing GPT-5.6 inside ChatGPT, swapped in for some customers who choose GPT-5.5 Professional.
Developer Anshu Chimala posted a side-by-side video on Thursday evaluating one-shot touchdown pages, captioning it: “Nicely effectively effectively, I am one of many fortunate ones with early GPT-5.6 Professional entry.”
Developer Dobroslav Radosavljevič posted on X that no matter he is working inside Codex, OpenAI’s coding agent, “feels waaaaaaaay totally different than [the] 5.5 mannequin.” Replies beneath cut up between believers and other people calling it placebo.
The clearest sample throughout the posts is time. Conor Dart, one of many many X customers amplifying the rumors, ran the check with a one-prompt 3D browser sport—with physics and digicam controls—that took simply over an hour to generate, versus GPT-5.5 Professional’s traditional 10-minute mark.
“Not good, however for a one-prompt AI sport dev check, that is significantly spectacular,” Dart wrote.
AI insider Chetas Lua reported an identical slowdown testing a robotic simulation, additionally fairly certain his outcomes come from OpenAI’s new mannequin: “GPT 5.6 Professional continues to mog [Anthropic’s Fable 5] in 3D check,” he wrote. “Engaged on video games one shot too.”
In a separate submit, he famous response instances stretching to twenty or 40 minutes, a tempo he mentioned hadn’t proven up since earlier than GPT-5.5 shipped.
Not each comparability flattered the rumored mannequin. Chris, an AI benchmarker on X, gave two fashions the identical spaceship-building immediate—the suspected GPT-5.6 Professional labored for 87 minutes in opposition to GPT-5.5 Additional Excessive’s 34 minutes and 42 seconds.
“As I’ve mentioned earlier than, based mostly on nice authority, GPT-5.6 will probably be an incremental/strong enchancment over GPT-5.5, not a Fable killer,” he wrote, whereas noting Fable 5 nonetheless beat each fashions on the spaceship’s core geometry. “My tough expectation has been that it will commerce blows with Fable 5 on some benchmarks, possibly win round half relying on the class, however not clearly surpass it total.”
A separate submit attributed to leaker Pankaj Kumar detailed leaks going additional: a information cutoff pushed to December 2025, a reasoning-effort setting some testers name “Juice Worth” allegedly raised to 960 from 768, and SVG and 3D design technology sturdy sufficient to beat Fable 5 on some duties.
None of it comes from OpenAI—however the particulars are constant throughout accounts: stronger reasoning, an unfinished entrance finish, and a launch candidate nicknamed Kindle-Alpha.
An AI influencer generally known as Leo, citing unnamed sources, wrote in a thread that the suspected mannequin is “now being stealth examined when 5.5 Professional is chosen in ChatGPT,” a minimum of for some Professional accounts, with a deliberate public launch the next Thursday, June 25.
The closest factor to an OpenAI fingerprint on any of this can be a memo, not a tweet. Chief scientist Jakub Pachocki reportedly instructed workers the following mannequin is a significant enchancment over GPT-5.5, in line with a report from The Data. That is nonetheless not affirmation of A/B testing, a launch date, or any spec floating round X, but it surely does affirm a brand new mannequin has been brewing.
Decrypt reached out to OpenAI to ask whether or not GPT-5.6 is being examined inside ChatGPT, however the firm didn’t reply by the point of publication.
Why OpenAI could be in a rush
If OpenAI is dashing a brand new flagship mannequin out the door, it has causes to. China’s open-source mannequin GLM-5.2 trails Claude Opus 4.8 by only one level on FrontierSWE—a benchmark that scores AI brokers on multi-hour, open-ended engineering initiatives by dominance charge—whereas beating GPT-5.5 outright on the identical check.
Anthropic, in the meantime, is coping with injury of its personal making. The corporate’s flagship Mythos 5 and Fable 5 fashions stay pulled underneath a U.S. export management directive issued June 12 over a disputed jailbreak vulnerability, leaving a spot on the prime of the market that GLM-5.2 and a hypothetical GPT-5.6 are each positioned to fill.
If Anthropic CEO Dario Amodei and President Donald Trump make peace, then Fable 5 can be rather more highly effective than another mannequin at present out there, with the standard hole between Anthropic’s prime mannequin and OpenAI’s changing into a lot larger than earlier than.
There’s additionally cash on the desk. OpenAI is reportedly weighing worth cuts on the tokens it expenses builders and enterprises, in line with the Wall Road Journal, in anticipation of Anthropic doing the identical as each corporations put together for dueling IPOs.
Whether or not any of it provides as much as an precise GPT-5.6 launch is one thing solely OpenAI can affirm, and the corporate has stayed quiet by every week of leaked checkpoints and stealth-testing claims. Polymarket merchants aren’t ready round, nevertheless—contracts on a launch between June 22 and June 28 have priced as excessive as 89% this week.
Day by day Debrief Publication
Begin day-after-day with the highest information tales proper now, plus unique options, a podcast, movies and extra.
