How to Deploy Kimi K2.6 Locally
Kimi K2.6 is a MoE model with 1,000B total parameters and 32B active per token. At Q4 it needs 559GB of VRAM to sit entirely on the GPU; with experts offloaded to system RAM it needs only 9GB of VRAM plus 552GB of RAM. No single card in the database fits it; it needs multiple GPUs or the unified memory of a Mac Studio. In offload mode the cheapest card that runs it is the Tesla V100 16GB, at roughly 4 tokens/s.
Kimi K2.6 details
How much VRAM does Kimi K2.6 need for local deployment?
MoE: all 1,000B parameters must stay resident in memory, while speed is set by the 32B active per token.
| Quantization | WeightsGB | Total at 8KGB | Total at 32KGB | Total at 128KGB |
|---|---|---|---|---|
| 1-bit | 234.7est. | 236.2 | 237.8 | 244.3 |
| 2-bit | 340.5GGUF | 373.1 | 374.7 | 381.2 |
| 3-bit | 479.2est. | 480.7 | 482.3 | 488.7 |
| 4-bit | 583.7GGUF | 558.9 | 560.5 | 567 |
| 5-bit | 674.7est. | 676.3 | 677.9 | 684.3 |
| 8-bit | 594.5GGUF | 1,038.1 | 1,039.7 | 1,046.1 |
| FP16 | 2,053.2GGUF | 1,957.3 | 1,958.9 | 1,965.4 |
Total = weights + KV cache (FP16) + 1GB runtime overhead. Weights marked GGUF are measured file sizes from the repository; the rest are estimated from bytes per parameter. Contexts beyond this model's 256K limit show a dash. Each tier takes whichever file of that bit width the repository has, preferring unsloth dynamic quants, so the exact file name varies by model.
Which GPUs can run Kimi K2.6?
Xianyu
No single card in the database fits Kimi K2.6: it needs at least 559GB of VRAM at Q4_K_M with an 8K context.
A multi-GPU build needs at least 559GB of VRAM in total.
None of the unified-memory machines in the database fit it either.
Recommended GPUs for Kimi K2.6 with CPU/GPU offloading
Xianyu
Expert weights live in system RAM (needs at least 552 GB); VRAM holds only attention layers, shared experts and the KV cache, at least 9 GB at Q4. Speed is bound by RAM bandwidth (estimated at 70 GB/s) and is far slower than a full-VRAM setup.
Deploy Kimi K2.6 locally with Ollama, llama.cpp or vLLM
The Ollama library has no official tag for it yet.
llama.cpp pulls the GGUF straight from Hugging Face; -c sets the context length.
llama-server -hf unsloth/Kimi-K2.6-GGUF:Q4_K_M -c 8192
vLLM serves the original-precision weights, which needs far more memory than GGUF and usually more than one GPU.
vllm serve moonshotai/Kimi-K2.6
File names follow that week's model-library snapshot; the definitions are in the methodology. Methodology
FAQ
How much VRAM does Kimi K2.6 need?
At Q4_K_M with an 8K context Kimi K2.6 needs about 559GB of VRAM; 672GB leaves comfortable headroom.
What is the cheapest GPU that runs Kimi K2.6?
No single card in the database fits Kimi K2.6; it needs multiple GPUs or a Mac Studio's unified memory.
Can you run Kimi K2.6 with Ollama?
Not yet: the Ollama library has no official tag for it. You can load the GGUF with llama.cpp instead.