Moonshot releases Kimi K3, a Chinese open-weight model with strong coding and efficiency

Moonshot has released Kimi K3, an open-weight model that has triggered a reassessment of how close Chinese models are to the frontier. The release is characterized by strong performance in coding, agentic tasks, and long-horizon knowledge work.

• Kimi K3 scores 57 on Artificial Analysis's Intelligence Index, placing it behind Claude Fable 5 (60) but ahead of Opus 4.8 (56).
• On coding agent benchmarks, it matches GPT-5.6 Terra and GPT-5.5, achieving 84% on Terminal-Bench v2 and 64% on DeepSWE.
• The model utilizes Kimi Delta Attention (KDA), a fast-weights memory mechanism that claims up to 6x faster throughput at 1M context length.
• K3 places China ahead of the US on Frontend Code Arena for the first time, debuting at #3 on DeepSWE as the first open-weight model with frontier-level results there.

The release shifts strategic focus from raw compute moats to efficiency stacks, suggesting that better post-training and infrastructure design can narrow capability gaps nonlinearly.