Chinas Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight AI model that the company says is the largest open-source model ever built. The model performs competitively with leading proprietary systems from Anthropic and OpenAI on several benchmarks, including frontend code generation where it beats Claude Fable 5.
Kimi K3 features a 1-million-token context window, native visual understanding, and an always-on reasoning mode. It is built on two architectural innovations: Kimi Delta Attention, a hybrid linear attention mechanism, and Attention Residuals, which the company describes as a drop-in replacement for residual connections that delivers consistent scaling gains.
The release sent shockwaves through US stock markets on Friday, with chip stocks declining as investors reassessed Americas AI lead over China. The model was trained despite US export controls limiting Chinese access to advanced semiconductors, demonstrating the effectiveness of Chinas workaround strategies.
Full model weights are scheduled for release on July 27. Early benchmark results show Kimi K3 matching or exceeding GPT-5.6 Sol on several reasoning and coding tasks while using fewer computational resources. As reported by TechCrunch and VentureBeat, the release marks a significant escalation in the global AI race and a validation of Chinas ability to compete in frontier AI development despite hardware restrictions.
