Just now, OpenAI has unveiled GPT-6. It is rumored that the model is equipped with 10 trillion parameters and will be forced to be released in August.
GPT-6 with 10 trillion parameters is set to be released later this month!
A bombshell news just broke out on X.
AI insider reporter ChrisGPT broke the news that despite the pressure of government censorship, GPT-6 (Astra) will still be released in August.
GPT-6 is the model Astra that OpenAI urgently suspended the release of yesterday.
Nine months ago, OpenAI stepped up its efforts and reallocated computing resources. Realizing that Anthropic was leading in pre-training, OpenAI grew anxious!
It is estimated that GPT-4 has about 1.8 trillion parameters. If Astra really has 10 trillion parameters, that is more than five times that of GPT-4.
Regardless, OpenAI has invested in an unprecedented scale in pre-training.
Don't rush, there are more powerful players coming later.
The second leak from ChrisGPT is the "epic behemoth" codenamed Doug, set to launch at the end of the year.
It is not only the largest pre-trained model OpenAI has ever built, but is also said to make Fable look like a primitive product.
And it is not GPT-6.
However, this model needs to go through pre-training first, followed by extensive cybersecurity testing, so it is expected to be released no later than November.
Coincidentally, recently, OpenAI President Greg reposted a post on X —
GPT-4 was trained on August 8, four years ago.
The industry speculates that GPT-4 has 1.8 trillion parameters. Four years later, GPT-6 has reached an astonishing 10 trillion parameters, completing a terrifying leap in order of magnitude!
It can be inferred that GPT-6 will exhibit unimaginable logical reasoning and generalization capabilities.
The Ultimate Behemoth Doug at the End of the Year:
Making Fable 5 Look Like a "Primitive Human"
The name Doug was not only revealed by ChrisGPT, but also previously mentioned by the well-known institution SemiAnalysis.
According to leaked information, Doug's computing power base is extremely top-tier.
Leading tech influencer Pankaj Kumar pointed out that Doug is most likely trained using Nvidia's next-generation Vera Rubin chips that have not yet been fully rolled out!
Overall, OpenAI is playing a big game.
In August, it will first use the 10-trillion-parameter GPT-6 to make a splash and attract users; by the end of the year, it will launch the Doug behemoth to claim the absolute top spot.
Tibo, who is in a fierce competition with Boris, also previewed a surprise
August, the Showdown for ASI
In July 2026, multiple sources revealed that OpenAI's upcoming next-generation model adopts a brand-new pre-training foundation (surpassing the previous "Spud" of about 4T tokens), with a scale of about 10T, a context window of up to 1.5M tokens, and is positioned to compete with Anthropic's Mythos/Fable.
The situation is now clear: Astra's release has been delayed due to safety pauses, but Anthropic has no intention of waiting.
Anthropic is preparing to release Fable 5.1 in August, priced the same as Fable 5, directly targeting Astra.
Anthropic is very likely to adopt the "Tian Ji's Horse Race" tactic: Opus 5 was deliberately underpowered at launch to use Fable 5.1 as a hidden troop, and will immediately intercept as soon as GPT-6 is unveiled.
Who will become the king of large models? This question has become more intense than ever in August.
Breaking the "Pre-training Wall": OpenAI's Jedi Counterattack
Why is Doug so important?
Because it points to a reality that the entire industry has been reluctant to face: since the release of GPT-4o in May 2024, OpenAI has not completed full-scale pre-training of its next-generation cutting-edge model for more than two years.
o1, o3, GPT-5 to GPT-5.5 are essentially post-training, reinforcement learning, and inference calculations built on the base of the GPT-4o era. GPT's "engine" has not been replaced for two years.
Due to severe pre-training bottlenecks, projects such as Orion have been downgraded due to underperformance.
However! OpenAI has never abandoned the core main line of pre-training.
The release of Gemini 3 at the end of last year completely convinced Sam Altman to replace the "engine".
Shortly afterwards, the new pre-trained model codenamed Garlic surfaced. It is reported that Chief Research Officer Mark Chen told the team that the company has solved the key problems in pre-training.
But Garlic is more like a verification experiment, and Doug is the real product after scaling up to a much larger scale.
They have likely overcome the performance degradation bottleneck and achieved stability for clusters of tens of thousands or even hundreds of thousands of GPUs, leading to the birth of Doug.
For a long time, SemiAnalysis has been emphasizing: Scaling Laws still work.
The o1 series proves that models can keep scaling at "inference time".
And the returning king Doug proclaims that in the basic dimension of "pre-training", as long as computing power and data keep up, Scaling Laws can still create miracles!
Only Two of the Three Giants Remain
Now, only two of the top three players are left.
On August 5, Google's AI territory suffered a major setback.
Jeff Dean, the chief scientist who has served for 27 years, left to start his own business together with three core researchers. On the same day, Nobel laureate Demis Hassabis stepped down as CEO of DeepMind and moved to the position of Chairman. Alphabet's stock price fell by more than 4% that day.
DeepMind is no longer a cutting-edge lab. The place where Transformer was invented is now sending top researchers to its competitors.
Not long ago, Silicon Valley still had the "three giants" of ASI, but now only OpenAI and Anthropic are left on the track.
Now Anthropic has Fable 5 in its hands. OpenAI has continuously iterated versions 5.5, 5.6, and the upcoming GPT-6.
Now OpenAI is holding two cards: Astra in August and Doug at the end of the year.
In the past two years, OpenAI has proven that the old engine can still squeeze out a lot of horsepower. When the engine is upgraded, how fast can this system, which has been tuned to its limits, push the vehicle to go?
The answer will be revealed by Doug at the end of the year.
References:
https://x.com/ChrisGPT/status/2086220662264250764
https://newsletter.semianalysis.com/p/gemini-is-cooked-but-gcp-is-cooking
This article is from the WeChat public account "AI Era", author: ASI Revelation, editor: Aeneas Solomon, authorized for release by 36Kr.