Beijing-based Moonshot AI released a new model on Friday that it says narrows the gap with top American systems, outperforming some of the most advanced offerings from OpenAI and Anthropic on specific benchmarks. The startup’s Kimi K3 model, which it claims is China’s largest AI system with 2.8 trillion parameters, beat Claude Opus 4.8 and GPT 5.5 on coding and general agent tasks. However, Moonshot acknowledged that K3 still trails the very best from US labs — Claude Fable 5 and GPT 5.6 Sol — on overall performance.

The release underscores the intensifying US-China competition in artificial intelligence, where Chinese models are gaining traction among Western companies for their lower cost and narrowing performance gap. Bank of America analysts led by Alex Liu noted that K3 shows pre-training scaling combined with architectural innovation can still deliver major improvements, even under China’s hardware and compute capacity constraints. Moonshot, founded in 2023, raised $2 billion at a valuation above $20 billion in May, with backing from Alibaba and Tencent.

The announcement sent shares of Chinese AI rivals tumbling. Z.ai, which launched a new model in June, saw its stock drop 28% on Friday, while MiniMax Group fell 16%. Alibaba, which also saw its stock decline 4% on Friday despite a boost earlier in the week from its Apple partnership in China, may face new pressure as Moonshot’s model challenges the open-source narrative around Alibaba’s Qwen series. Liu added that K3 raises the capability ceiling for Chinese AI models, shifting the burden of proof to other independent labs.

Source: CNBC