Open-Source AI

By early 2025, the AI world had a new "Sputnik moment." DeepSeek-R1  had just exploded onto the scene, proving that a lean team from Hangzhou could deliver GPT-4 level reasoning for a fraction of the cost. As detailed in the DeepSeek-R1 Paper , the model incentivized reasoning through Reinforcement Learning rather than just raw scale. It was the ultimate "catfish" in the tank, forcing Western titans to defend their multi-billion dollar "brute force" scaling strategies. Now, a

While Silicon Valley was fixated on Palo Alto, a quiet revolution was brewing 6,000 miles to the West. In January 2026, MiniMax Group Inc. (SEHK: 0100)  made history as the world’s fastest AI startup to reach a public listing, debuting on the Hong Kong Stock Exchange  just four years after its founding. Unlike many of its peers, MiniMax isn't just chasing technical benchmarks—it’s chasing users, and it’s winning. With over 70% of its revenue  now originating from overseas mar

GLM-4.7 vs GPT-5.2 coding benchmark While the world spent the last five years staring at Palo Alto, the tectonic plates of Artificial Intelligence just shifted 6,000 miles to the West. It is 2026, and the "Intelligence Gap" has officially closed. In a single week, the narrative of Silicon Valley’s untouchable dominance was shattered—not by a better marketing campaign, but by a 94.2% math reasoning score and a $14 billion IPO. Meet Zhipu AI , the Tsinghua-born powerhouse that

Prologue: The Memory Bottleneck In the race for Artificial General Intelligence (AGI), Silicon Valley has traditionally been obsessed with compute —the sheer brute force of reasoning. But in Beijing, a rising star named Moonshot AI  identified a different, more human bottleneck: Memory. As founder Yang Zhilin famously posed: "If an AI cannot remember an entire book, how can it truly understand and reason about a complex world?" While competitors were busy building flashy avat

Prologue: The Heist That Wasn't Silicon Valley, December 2025. The air is thick with a new kind of tension. In venture capital meetings, developer forums, and the halls of Washington D.C., a single name is whispered: "Qwen." The story begins not with a typical product launch, but with a quiet, seismic release on a developer platform. It ends with a U.S. government scramble to understand an entirely new policy problem: how to regulate an open-source AI model anyone could downl