No — not on this device
Kimi K2.6 1T-A32B at Q4_K_M needs 578.6 GB against 27.0 GB usable, and the shortfall of 551.6 GB is more than 32 GB of system RAM can cover at a tolerable speed. A smaller sibling or a lower quantisation is the honest answer here.
Does not fit
8K context
Q4_K_M · 577.5 GB
Modified MIT
Released 20 Apr 2026
Not in the Ollama library
The open coding-agent benchmark leader of spring 2026 (80.2 SWE-bench). A 512 GB Mac Studio pair, or a ceiling.
The VRAM budget
weights 577.5 GB
Weights 577.5 GB
KV cache @ 8K 0.54 GB
Runtime overhead 0.6 GB
Over budget 551.6 GB past 27.0 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 1016.2 GB | 1017.4 GB | — | ~1.2 | −0.1% ppl | 990.4 GB over |
| Q6_K | 784.3 GB | 785.4 GB | — | ~1.6 | −0.4% ppl | 758.4 GB over |
| Q5_K_M | 677.9 GB | 679.0 GB | — | ~1.9 | −0.8% ppl | 652.0 GB over |
| Q4_K_M | 577.5 GB | 578.6 GB | — | ~2.2 | −1.9% ppl | 551.6 GB over |
| Q3_K_M | 467.5 GB | 468.6 GB | — | ~2.7 | −5.4% ppl | 441.6 GB over |
| Q2_K | 400.5 GB | 401.7 GB | — | ~3.1 | −15% ppl | 374.7 GB over |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. This model uses multi-head latent attention, so its cache is a compressed latent rather than full K and V.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/Kimi-K2.6-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.
01Download is 577.5 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02macOS caps what the GPU may wire down at about 27.0 GB of 36 GB. Raising it with iogpu.wired_limit_mb is possible, and risky.
03If it falls back to CPU silently, drop the context first, then step down a quantisation.