It's the first time in the history of the world.
The 10 trillion-parameter GPT-6 will be released this month!
Just as X was uploaded, news broke.
AI insider Chris GPT reported that despite government censorship pressures, GPT-6 (Astra) will be released in August.
The GPT-6 is a model of the Astra that OpenAI released yesterday with an emergency suspension.
Nine months ago, OpenAI began to reinforce and redistribute computing resources. Realizing that Anthropic was leading the way in pre-training, OpenAI was in a hurry!
It is estimated that the parameters of GPT-4 are about 1.8 trillion. If the Astra parameters are actually 10 trillion, that is more than five times the GPT-4 parameters.
In any case,OpenAI has emerged at an unprecedented scale in pre-training.
Don't hurry, there's a gentleman in the back.
Chris GPT's second release was the year-end "Epic Monster" by Doug.
Not only is it the largest pre-training model of OpenAI to date, but it is also said to make Fable look like an original product.
And it's not the GPT-6.
However, the model will need to be pre-trained and then extensive cybersecurity testing, so it is likely to be released by November at the latest.
Incidentally, just recently, Greg, the president of OpenAI, sent out an X-ray.
GPT-4 was completed on August 8, four years ago.
Industry speculates that the parameters of GPT-4 are at 1.8 trillion, and four years later, the parameters of GPT-6 are at 1 trillion, completing a terrifying leap in quantity!
It can be speculated that GPT-6 will emerge with unimaginable logical reasoning and generalization capabilities.
The ultimate monster at the end of the year, Doug:"Fable 5" is "original"Doug's name was not only revealed by Chris GPT, but also by the well-known SemiAnalysis.
According to reports, Doug's calculating base is extremely luxurious.
V Pankaj Kumar pointed out that Doug was most likely trained using the next generation of Vera Rubin chips that had not yet been fully developed by Ingrid!
Overall, OpenAI is in the next big game.
In August, the 10 trillion-dollar GPT-6 exploded into the ground, attracting users; by the end of the year, Doug, the giant, had won the absolute first throne.
Tibo, who fought against Boris, also predicted surprises.
August, the final battle of the ASIIn July 2026, it was widely reported that the next generation of OpenAI models would be based on a completely new pre-training base, exceeding the previous 4T token called "Spud" by about 10T, and could reach 1.5M tokens in the context of Anthropic's Mythos/Fable.
The situation is now clear: Astra's release date was delayed due to security breaks, but Anthropic did not intend to wait.
Anthropic is preparing to release Fable 5.1 in August, priced at the same level as Fable 5, directly against Astra.
Anthropic is most likely to use the "Taco Horse" tactic: the intention of Opus 5 is to use Fable 5.1 as an ambush when the GPT-6 makes an immediate impact.
Who will be the King of the Big Model? The question became unprecedented in August.Breaking through the "Pre-Training Wall": OpenAI is definitely back
Why is Doug so important?
Because it points to a reality that the entire industry has not dared to face: OpenAI has not completed full-scale pre-training of the next generation of front-end models for more than two years since the release of GPT-4o in May 2024.
o1, o3, GPT-5 to GPT-5.5, essentially the backbone of the GPT-4o era of post-training, reinforcement learning and computational reasoning. GPT's "engine" has remained unchanged for two years.
Due to severe pre-training bottlenecks, projects such as Orion were downgraded due to poor performance.
However, OpenAI has never abandoned this core core of pre-training.
The Gemini 3, which was released at the end of last year, completely changed Ottman's decision to change the "engine".
Soon after, a new pre-training model, code-named Garlic, surfaced. Mark Chen, chief research officer, reportedly told the team that the company had solved a key problem in pre-training.
But Garlic is more like a proof-of-concept experiment, and Doug is the real product of a larger scale.
It is likely that they overcame the bottleneck of performance degradation and established the stability of the Wanka or even the 100,000-card cluster that led to the birth of Doug.
For a long time, SemiAnalysis has emphasized that scaling laws remain valid.
O1 series demonstrates that models can be continuously scaled over "reasonable time".
When the King returned, Doug declared: "On the basis of the 'pre-training' dimension, Scaling Laws can still create miracles as long as computation and data keep up!"
Three giants, only two.Today, there are only two left of the three.
On August 5, Google's AI chart collapsed.
For 27 years, Chief Scientist Jeff Dean and three core researchers set out to start a startup. On the same day, the Nobel Prize winner Hasabis stepped down as CEO of DeepMind and retired as chairman. Alphabet shares fell more than 4% on the same day.
DeepMind, which is no longer a cutting-edge laboratory -- the place where the Transformer was invented -- is sending top researchers to rivals.
Not long ago, Silicon Valley was the "Three Giants" of ASI, and now only OpenAI and Anthropic remain on the track.
Today, the Anthropic hand is Fable 5. The OpenAI hand is a continuous iteration of 5.5, 5.6, and the upcoming 6.
Now, OpenAI is holding two cards: August's Astra and the year-end's Doug.
Over the past two years, OpenAI has proven that old engines can still produce a lot of horsepower.
This is the answer, which Doug will reveal by the end of the year.