What hardware runs Kimi K2.5?
Kimi K2.5 is a 1026.41B-parameter mixture-of-experts model with 32.86B active parameters (other license). A trillion parameters, 33B active. The strongest open agentic coder going — and the definition of rent-don't-buy.
All figures assume an f16 KV cache, a 0.6 GB display reserve on the GPU, and 64 GB of DDR5 system RAM for the offload tiers. Tune these in the calculator.
Minimum VRAM by quant and context
| Quant | File size | 8K | 16K | 32K | 64K | 128K |
|---|---|---|---|---|---|---|
| Q2_K * | 348.1 GB | 351.2 GB | 352.5 GB | 353.8 GB | 355.9 GB | 360.2 GB |
| Q3_K_M | 456.1 GB | 459.2 GB | 460.5 GB | 461.8 GB | 464 GB | 468.3 GB |
| Q4_K_M | 578.6 GB | 581.7 GB | 583 GB | 584.3 GB | 586.4 GB | 590.7 GB |
| Q5_K_M | 678.7 GB | 681.8 GB | 683.1 GB | 684.4 GB | 686.5 GB | 690.8 GB |
| Q6_K | 785.0 GB | 788.1 GB | 789.4 GB | 790.7 GB | 792.9 GB | 797.1 GB |
| Q8_0 | 1016.1 GB | 1019.2 GB | 1020.5 GB | 1021.8 GB | 1024 GB | 1028.3 GB |
| IQ4_XS | 509.6 GB | 512.7 GB | 514 GB | 515.3 GB | 517.4 GB | 521.7 GB |
Full-GPU figures: weights + f16 KV cache + overhead. * below our recommended floor of Q3_K_M.
GPU compatibility
Fits on GPUExpert offloadPartial offloadCPU only
Quant guidance
Our floor for Kimi K2.5 is Q3_K_M — below that, quality degrades faster than the VRAM savings are worth (large models tolerate low bpw better than small ones, which is why the floor is lower here). Prefer the highest quant that still lands "Fits on GPU" or “Expert offload” at your context length in the table above.
Frequently asked questions
How much VRAM does Kimi K2.5 need?
At Q3_K_M and 32K context, Kimi K2.5 needs about 461.8 GB of VRAM to run fully on GPU (weights + KV cache + overhead). As a mixture-of-experts model it can also run with far less VRAM via expert offload, keeping experts in system RAM.
What is the smallest GPU that can run Kimi K2.5?
No single consumer GPU in our set runs Kimi K2.5 well — it's a cloud-rental or multi-GPU model.
What quantization should I use for Kimi K2.5?
We recommend Q3_K_M or higher. Q4_K_M weighs 578.6 GB (4.84 bits/weight); going below Q3_K_M costs noticeable quality on a model this size.
How long a context can Kimi K2.5 handle?
Kimi K2.5 supports up to 256K tokens. KV cache grows linearly with context, though its MLA attention compresses the KV cache dramatically.
Can I run Kimi K2.5 without a GPU?
Not realistically — even at 8K context the weights don't fit in 64 GB of system RAM.