LLM EXPLORER 57,918 MODELS INDEXED

Kimi K3 Mlx Reap160 2bit by mlx-community

By mlx-community · 998 downloads
HF Repo ↗ Copy URL

Kimi K3 Mlx Reap160 2bit is an open-source language model by mlx-community. Features: 50.2b LLM, VRAM: 181.7GB, License: other, Quantized, LLM Explorer Score: 0.46, ELO: 1485.

  Arxiv:2510.13999   2bit   Apple-silicon Base model:finetune:moonshotai...   Base model:moonshotai/kimi-k3   Code   Custom code   Kimi-k3   Kimi k3   Mlx   Moe   Pruned   Quantized   Reap   Region:us   Safetensors   Sharded   Tensorflow
Model Card on HF 🤗: https://huggingface.co/mlx-community/Kimi-K3-mlx-reap160-2bit 

Kimi K3 Mlx Reap160 2bit Benchmarks

LMSys ELO 1485 vs 1272 (so35) 16.7%
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

Kimi K3 Mlx Reap160 2bit Parameters and Internals

LLM NameKimi K3 Mlx Reap160 2bit
Repository 🤗https://huggingface.co/mlx-community/Kimi-K3-mlx-reap160-2bit 
Base Model(s)  Kimi K3   moonshotai/Kimi-K3
Model Size50.2b
Required VRAM181.7 GB
Updated2026-08-10
Maintainermlx-community
Model Typekimi_k3
Model Files  5.7 GB: 1-of-33   5.8 GB: 2-of-33   5.8 GB: 3-of-33   5.8 GB: 4-of-33   5.4 GB: 5-of-33   5.8 GB: 6-of-33   5.8 GB: 7-of-33   5.8 GB: 8-of-33   5.4 GB: 9-of-33   5.8 GB: 10-of-33   5.8 GB: 11-of-33   5.8 GB: 12-of-33   5.4 GB: 13-of-33   5.4 GB: 14-of-33   5.7 GB: 15-of-33   5.4 GB: 16-of-33   5.8 GB: 17-of-33   5.8 GB: 18-of-33   5.8 GB: 19-of-33   5.4 GB: 20-of-33   5.8 GB: 21-of-33   5.8 GB: 22-of-33   5.8 GB: 23-of-33   5.4 GB: 24-of-33   5.4 GB: 25-of-33   5.7 GB: 26-of-33   5.4 GB: 27-of-33   5.8 GB: 28-of-33   5.8 GB: 29-of-33   5.8 GB: 30-of-33   5.4 GB: 31-of-33   5.4 GB: 32-of-33   0.8 GB: 33-of-33
Quantization Type2bit
Model ArchitectureKimiK3ForConditionalGeneration
Licenseother