Kimi K3.1 Pro — Weights Released Exactly as Promised; Independently Verified Top Open-Weight, with Two Real Limitations
Full checkpoint: 2.8T total / 104B active (MoE, 896 experts, 16+2 shared per token), 93 layers (69 KDA + 24 Gated DSA — replacing K3's full-attention MLA, cheaper for long context), 1,048,576-token context, multimodal (MoonViT-V2 401M). Weights are **MXFP4 with MXFP8 activations, quantization-a…
Full checkpoint: 2.8T total / 104B active (MoE, 896 experts, 16+2 shared per token), 93 layers (69 KDA + 24 Gated DSA — replacing K3's full-attention MLA, cheaper for long context), 1,048,576-token context, multimodal (MoonViT-V2 401M).
Weights are **MXFP4 with MXFP8 activations, quantization-aware trained** — the precision Moonshot itself serves, not an after-the-fact compression; 96 shards, ~1.56 TB.
It is treated as a post-training of K3.
Independently verified: ArtificialWatch's aggregate index places it **#11 of 44 models (69.9) — the highest-ranked model with downloadable weights**, displ…