FreeLLMMoonshot AI launches Kimi‑K3.1‑Pro, a 2.8‑trillion‑parameter multimodal model on Hugging Face
The new Kimi‑K3.1‑Pro model adds a 1‑million‑token context window and shows strong results on coding, 3D and finance benchmarks.
The model is open‑weight and multimodal, built on a 2.8 trillion‑parameter Mixture‑of‑Experts architecture with 104 billion activated parameters.
It supports a 1 million‑token context length and uses a Vision Encoder called MoonViT‑V2. Quantization is MXFP4 for weights and MXFP8 for activations.
Benchmarks show Kimi‑K3.1‑Pro leading in several tasks: 92.7 % on BenchCAD with tools, 77.1 on Kimi Code Bench 2.0, and 95.4 % on BabyVision. It also scores 66.4 on Terminal‑Bench coding, close to DeepSeek‑V4.1‑Flash‑1017’s 66.4.
FreeLLMArena’s latest blind vote ranks the earlier Kimi K3 model as the top free chat model, indicating strong community interest in Moonshot’s releases.
あなたにとっての意味
Kimi‑K3.1‑Pro offers a high‑capacity, open‑weight option for developers needing extensive context and multimodal capabilities, expanding the free‑model ecosystem.
ソース
- Moonshot AI releases Kimi-K3.1-Pro on Hugging Face, Hugging Face trending (official), 2 Nov 2026
上記のソースから自動的に生成し、当社のコードでチェックしました：すべての数値と名前が含まれ、文がソースをコピーしてはいけません。間違いを見つけましたか？報告する。
今日無料で使えるAIはどれですか？
すべての無料AIモデルをブラインドヒューマン投票で順位付けし、あなたのGPUで必要なメモリと速度を示します。