Skip to content
localmodel.run

Model family · 3 sizes

Kimi: which size runs locally?

Kimi comes in 3 sizes, from 1000B. Some sizes are Mixture-of-Experts, so they run faster than their memory footprint suggests. Here is each size with its Q4_K_M weight, the memory it needs, and the hardware that runs it.

Sizes
3
Smallest
1000B
Largest
1000B

The Kimi lineup

Smallest first Q4_K_M · min memory

"Needs" is the sourced minimum memory for Q4_K_M with a small context. Larger context needs more.

Which Kimi fits your memory

8GB

No Kimi size fits 8GB; even Kimi K2 Instruct needs more.

No
16GB

No Kimi size fits 16GB; even Kimi K2 Instruct needs more.

No
24GB

No Kimi size fits 24GB; even Kimi K2 Instruct needs more.

No
32GB

No Kimi size fits 32GB; even Kimi K2 Instruct needs more.

No
48GB

No Kimi size fits 48GB; even Kimi K2 Instruct needs more.

No
64GB

No Kimi size fits 64GB; even Kimi K2 Instruct needs more.

No
128GB

No Kimi size fits 128GB; even Kimi K2 Instruct needs more.

No
256GB

No Kimi size fits 256GB; even Kimi K2 Instruct needs more.

No

Best case means the most capable device at that size (usually a discrete GPU). A Mac at the same size sits roughly one rung lower; see the per-size breakdown on each memory budget page.

FAQ

Which Kimi size should I run locally?

Pick the largest size your memory allows. . Smaller sizes run faster and leave headroom for context.

What is the smallest Kimi model?

Kimi K2 Instruct at 1000B parameters, about 578.15 GB on disk at Q4_K_M. It is the one to use on phones and 8 GB machines.

What is the largest Kimi model and what does it need?

Kimi K2.7 Code at 1000B (mixture of experts), about 583.71 GB at Q4_K_M. It needs more than a typical 32 GB desktop; a high-memory Mac or multi-GPU rig.

Understand the numbers

Short guides to the ideas behind Kimi's memory and quant figures.

Sources

Memory figures are estimates at Q4_K_M. See methodology.