Chinese AI startup Moonshot AI has launched Kimi K3, an open-weight large language model with a staggering 2.8 trillion parameters, claiming it to be the world's largest of its kind and the first to approach the 3 trillion mark. The release instantly sent shockwaves through the Chinese AI sector, with rivals Zhipu and Minimax seeing their shares plunge 27.7% and 16.5% in Hong Kong trading.
Kimi K3's 1 million-token context window and architectural innovations — including Kimi Delta Attention for faster decoding and Attention Residuals for training efficiency — helped it top UC Berkeley's Arena coding leaderboard and land second overall on Vals AI benchmarks, just behind Anthropic's Fable 5. Moonshot claims the model performs competitively with frontier US systems on complex reasoning and long-horizon coding tasks, though it acknowledged it still trails the most powerful proprietary models from Anthropic and OpenAI.
The launch drew comparisons to DeepSeek's disruptive 2025 debut, with tech investor Kevin Xu calling it a "violent market reaction" and University of Pennsylvania professor Ethan Mollick deeming it "closest to the frontier yet" among Chinese models. Analysts note that open-weight models like Kimi K3 can be run at a fraction of the cost of closed-source services, potentially accelerating global AI adoption — a development that could reverberate across crypto and blockchain projects relying on AI integrations.
Moonshot, backed by Alibaba and Tencent, is reportedly seeking $2 billion in funding at a $30 billion valuation ahead of a possible Hong Kong listing. The model's release coincides with U.S. government delays on public rollouts from OpenAI and Anthropic due to security concerns, further spotlighting the intensifying AI race and its implications for both traditional and digital asset markets.
