Skip to main content

GPT-6 Leak: 10 Trillion Parameters, August Launch, and the Mystery of 'Doug'

The AI rumor mill is churning, and this time the whispers are loud enough to wake the whole industry. According to AI insider ChrisGPT, OpenAI is gearing up to release GPT-6—the model previously code-named Astra—in August, despite government pushback. If the numbers are right, we're looking at a beast with around 10 trillion parameters, which would make it more than five times larger than GPT-4. That's not just an incremental step; that's a leap across the pre-training chasm.

But here's the kicker: GPT-6 might just be the opening act. ChrisGPT also teased an upcoming "epic monster" called Doug, reportedly the largest pre-trained model in OpenAI's history. Doug isn't GPT-6. It's something else entirely, likely built on NVIDIA's next-generation Vera Rubin chips, which haven't even hit the market yet. With pre-training and cybersecurity testing still ahead, Doug is expected to land by November at the latest. The strategy seems clear: GPT-6 grabs attention in August, and Doug swoops in by year-end to reclaim absolute leadership.

The Pre-Training Wall Finally Cracks

Doug matters because it signals a long-awaited breakthrough. Since GPT-4o dropped in May 2024, OpenAI hasn't completed a full-scale pre-training run for its next-gen flagship. Models like o1, o3, GPT-5, and GPT-5.5 were essentially post-training and reinforcement learning layered on top of the GPT-4o foundation. The engine under the hood hasn't been swapped in over two years. Projects like Orion were quietly downgraded when they underperformed. But then Gemini 3 shook things up last year, and the boss decided enough was enough. A verification experiment called Garlic proved the concept, and Doug is the real deal after scaling up. This suggests OpenAI may have cracked the performance degradation and stability issues that plague ten-thousand-card clusters, proving that Scaling Laws still hold in pre-training.

The Competitive Landscape Shifts

Meanwhile, the chessboard is changing fast. On August 5th, Google's chief scientist Jeff Dean left with three core researchers to start his own venture. Hassabis stepped down as DeepMind CEO to become chairman, and Alphabet's stock took a 4% hit that day. The birthplace of the Transformer is now exporting its brightest minds to the competition. The so-called "three giants" of Silicon Valley ASI are down to two: OpenAI and Anthropic.

Anthropic isn't sitting idle. They're holding Fable 5 and preparing to release Fable 5.1 in August, keeping the price steady and directly targeting Astra. It's a classic "Tian Ji's horse racing" strategy—sacrifice a weaker piece to save the stronger Opus 5 for later. OpenAI, on the other hand, is juggling both GPT-6 and Doug. The old engine has been pushed to its limits over the past two years. When the new engine finally kicks in, how fast can this system run? Doug will answer that question by year-end.

Key Points

  • GPT-6 is set to launch in August with ~10 trillion parameters, five times larger than GPT-4.
  • Doug, a mysterious model, is expected by November, possibly using NVIDIA's Vera Rubin chips.
  • OpenAI hasn't done full-scale pre-training since GPT-4o, but Doug marks a return to engine upgrades.
  • Google's talent exodus and Anthropic's strategic moves are reshaping the AI race.
  • The industry is watching to see if Scaling Laws still hold in pre-training.