Moonshot AI
Kimi Linear 48B A3B (48B parameters) requires approximately 33.2 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 39 GB of VRAM.
Get started
— copy & paste to run locallyCopy-paste commands to run Kimi Linear 48B A3B on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "moonshotai/Kimi-Linear-48B-A3B-Instruct" \
--hf-file "Kimi-Linear-48B-A3B-Instruct-Q4_K_M.gguf" \
-c 4096 -ngl 99Quick specs
About this model
Quick picks
Best hardware
Run this model
Quantization options
No hardware detected — fit column shows raw VRAM estimates
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 18.7 GB | Low | — |
Q3_K_S | 3 | 23.5 GB | Low | — |
NVFP4 | 4 | 26.9 GB | Medium | — |
Q4_K_M | 4 | 29.3 GB | Medium | — |
Q5_K_M | 5 | 34.6 GB | High | — |
Q6_K | 6 | 39.4 GB | High | — |
Q8_0 | 8 | 51.4 GB | Very High | — |
F16 | 16 | 98.4 GB | Maximum | — |
Quality benchmarks
Reasoning
Source: official · 2025-10-30
Hardware compatibility
Computing compatibility...
Memory breakdown
Frequently asked questions
Kimi Linear 48B A3B (48B parameters) requires approximately 33.2 GB of VRAM with Q4_K_M quantization. Lower quantizations like Q4_K_M use less memory but may reduce quality.
Yes, Mac mini M4 64GB can run Kimi Linear 48B A3B with a compatibility score of 77/100. It provides 64 GB of memory and achieves approximately 5.3 tokens per second.
The recommended quantization for Kimi Linear 48B A3B is Q4_K_M, which offers the best balance between model quality and memory efficiency. Higher quantizations preserve more quality but require more VRAM.
The top recommended hardware for Kimi Linear 48B A3B: RTX PRO 5000 Blackwell 48GB (score: 86/100), RTX 6000 Ada 48GB (score: 85/100), NVIDIA H100 80GB (score: 85/100). These provide the best combination of memory, bandwidth, and compute for running this model locally.
Yes, Kimi Linear 48B A3B is well-suited for chat as well as reasoning, long-context, coding. It was designed with these use cases in mind.
See also
