Kimi K3.1 Pro Takes Aim at GPT-6

🔥 What happened Moonshot AI dropped Kimi K3.1 Pro: an open-weight 2.8T-parameter MoE with a 1M-token context and native multimodality. It swaps full attention for Gated DeepSeek Sparse Attention and beats rivals across several agentic benchmarks. 💡 Why it matters With just 104B activated parameters, it hits 70.0 on HLE-Full with tools — ahead of GPT-6 Astra (57.2) and Claude Fable 5.1 (65.6). Open weights plus MXFP4 quantization make frontier-grade agents realistic on your own hardware for the first time. ⚡ Our take The benchmark table is a battlefield, and Kimi wins more columns than it loses. Anyone still calling closed models the only option hasn't done the math.