Chinese startup Moonshot AI has released Kimi K3, a 2.8-trillion-parameter system that the company describes as the world's largest open-weight artificial intelligence model.
According to Tom's Hardware, the Beijing-based firm calls Kimi K3 the world's first open "3T-class" system, and reports that it beat Anthropic's Claude Fable 5 in the Frontend Code Arena benchmark. Tom's Hardware frames the release as evidence that Chinese developers are working around U.S. restrictions on advanced computing chips.
Moonshot says Kimi K3 matches or outperforms leading proprietary models on several coding and software-engineering benchmarks, according to Business Standard, which points to the release as a sign of the rising strength of open-weight AI.
Reuters reports that the launch shows China's AI sector closing in on U.S. rivals, while oodaloop.com and Yahoo Tech cast the model as a direct challenge to OpenAI and Anthropic, with the performance gap narrowing quickly. The Star reports that Moonshot is chasing a "DeepSeek moment" — a reference to the earlier Chinese model that rattled the industry — with its much-hyped release.
The key distinction is "open-weight": unlike the closed systems sold by OpenAI and Anthropic, an open model's underlying parameters are made available, letting outside developers download, run, and build on it more freely. That openness, combined with the model's record size, is what has drawn attention.
Why it matters: if a Chinese lab can ship the largest open AI model despite U.S. chip controls and rival the best American systems on coding tasks, it signals that the global AI race is tightening — and that cutting-edge capability is spreading beyond a handful of closed, U.S.-based providers.