AI insider reporter ChrisGPT revealed that despite government censorship, GPT-6 (previously temporarily paused Astra model) will be released in August. It is estimated that GPT-4 has about 1.8 trillion parameters, and if Astra reaches 10 trillion parameters, it would be more than five times larger than GPT-4. OpenAI has made an unprecedented scale in pre-training. Just nine months ago, OpenAI realized that Anthropic was leading in pre-training and urgently reallocated computing resources. Now the time has finally come.

image.png

However, GPT-6 in August may just be a starter. ChrisGPT also disclosed an upcoming "epic monster" called Doug, which is said to be the largest pre-trained model in OpenAI's history, making the flagship Fable seem like a primitive product. Doug is not GPT-6. According to insiders, it is likely to use the next-generation Vera Rubin chips from NVIDIA, which have not been fully rolled out yet. Due to the need for pre-training and extensive cybersecurity testing, it is expected to be released by November at the latest. OpenAI's strategy is clear: GPT-6 will attract users in August, and then Doug will be released at the end of the year to regain absolute leadership.

The wall of pre-training has finally been broken, and the rivalry between two major players is established

Doug is crucial because it points to a reality the industry has been reluctant to face: since the release of GPT-4o in May 2024, OpenAI has not completed full-scale pre-training for its next-generation cutting-edge model for over two years. o1, o3, GPT-5, and GPT-5.5 are essentially post-training and reinforcement learning based on the GPT-4o foundation. The "engine" of GPT has not been replaced for two years. Projects like Orion were downgraded due to unsatisfactory performance, until the release of Gemini 3 last year prompted the boss to decide to replace the engine. The verification experiment called Garlic was completed first, and Doug is the true product after scaling up. This means that OpenAI may have overcome the performance degradation and stability bottlenecks of ten thousand card clusters, proving that Scaling Laws still hold in the pre-training dimension.

At the same time, the competitive landscape has changed dramatically. On August 5th, Google's chief scientist Jeff Dean left with three core researchers to start his own business. Hassabis stepped down as CEO of DeepMind and moved to chairman. Alphabet's stock dropped over 4% that day. The place where the Transformer was invented is now supplying top talent to competitors. The "three giants" of Silicon Valley ASI now only have OpenAI and Anthropic left.

Anthropic holds Fable 5 and is preparing to release Fable 5.1 in August, keeping the price unchanged and directly targeting Astra. It seems to be using a "Tian Ji's horse racing" strategy, saving the strength of Opus 5 and launching Fable 5.1 as an ambush. OpenAI is holding both GPT-6 and Doug. The old engine has been squeezed to the limit over the past two years. When the engine is truly upgraded, how fast this system can run will be answered by Doug at the end of the year.