Kimi K3: Moonshot AI's 2.8-Trillion-Parameter Open Frontier Model — Benchmarks, Architecture, and Everything We Know

Short summary

Moonshot AI launched Kimi K3, a 2.8-trillion-parameter MoE model with a 1M-token context window, native vision, and novel KDA and AttnRes architecture. It benchmarks competitively against Claude Fable 5 and GPT-5.6 Sol at roughly half the price, leading on several coding and agentic benchmarks. Full open weights are promised by July 27, 2026.

• •2.8T-parameter MoE model with 1M-token context and native vision
• •Novel KDA and AttnRes architecture cuts KV-cache 75% and boosts training efficiency 25%
• •Competitive benchmarks vs Claude Fable 5 and GPT-5.6 Sol at half the price; open weights promised by July 27

Generated with AI, which can make mistakes.