Moonshot's Kimi K3 Is the Largest Open-Weight Model Yet, and It Rivals the US Labs

Moonshot AI, the Beijing startup backed by Alibaba, released Kimi K3, a 2.8-trillion-parameter open-weight model that it calls the largest open-source model released to date, per CNBC. On benchmarks it lands close to the top proprietary systems: it trails Claude Fable 5 and GPT 5.6 Sol overall but beat Claude Opus 4.8 and GPT 5.5 on coding and agent tasks, per VentureBeat. The release was timed to land just ahead of the World Artificial Intelligence Conference in Shanghai.

The detail that matters is "open weight at 2.8 trillion parameters." The weights are downloadable, so any lab or enterprise can self-host and fine-tune rather than rent through an API. And in blind testing by the Arena evaluator, developers preferred Kimi over every leading US model for front-end coding, per Axios. That is revealed preference from the people who actually write the code, which carries more weight than any benchmark table. A frontier-adjacent model you can download and run, priced well below the US premium tier, changes the build-versus-buy math for anyone shipping AI features.

This extends a pattern we have tracked. Xiaomi's open-weight models already lead OpenRouter usage, Anthropic cut Sonnet pricing in response, and Kimi K3 now pushes open weights to near-frontier capability. The commodity middle of the model market is being competed toward zero while the US labs keep only the very top. For pricing power, that ceiling keeps dropping.