Moonshot AI released Kimi K3, a 2.8 trillion parameter model. It is set to become one of the largest open-weight models when full weights drop by July 27. Internal tests and independent evaluations show it performing close to or beating Claude Fable 5 and GPT-5.6 Sol on several tasks. It leads on Arena.ai’s front-end development leaderboard. The model excels in coding, agentic work, and real-world evaluations while trailing slightly overall. This release comes amid US restrictions on advanced AI and concerns over model distillation. Moonshot’s progress challenges assumptions about China lagging behind American labs. It follows earlier moves like DeepSeek and raises questions about export controls and national security. Analysts see it as a significant moment for open-source AI from China.
Moonshot AI released Kimi K3, a 2.8 trillion parameter model. It is set to become one of the largest open-weight models when full weights drop by July 27. Internal tests and independent evaluations show it performing close to or beating Claude Fable 5 and GPT-5.6 Sol on several tasks. It leads on Arena.ai’s front-end development leaderboard. The model excels in coding, agentic work, and real-world evaluations while trailing slightly overall. This release comes amid US restrictions on advanced AI and concerns over model distillation. Moonshot’s progress challenges assumptions about China lagging behind American labs. It follows earlier moves like DeepSeek and raises questions about export controls and national security. Analysts see it as a significant moment for open-source AI from China.