FitMyLLM
Moonshot AI/Mixture of Experts

MKimi K2.5

Moonshot Kimi K2.5 — MoE model with strong reasoning and tool use capabilities.

chatcodingreasoningvisiontool_useThinkingTool Use
1000B
Parameters (32B active)
256K
Context length
13
Benchmarks
17
Quantizations
3.4M
HF downloads
Architecture
MoE
Released
2026-02-02
Layers
61
KV Heads
64
Head Dim
128
Family
kimi

Quantization Options

QuantBitsVRAMQuality
IQ2_XXS2.38298.0 GBlow
IQ2_M2.93366.7 GBlow
Q2_K3.16395.5 GBlow
IQ3_XXS3.25406.7 GBlow
IQ3_XS3.5438.0 GBlow
Q3_K_S3.64455.5 GBlow
IQ3_M3.76470.5 GBlow
Q3_K_M4500.5 GBlow
Q3_K_L4.3538.0 GBmoderate
IQ4_XS4.46558.0 GBmoderate
Q4_K_S4.67584.2 GBmoderate
Q4_K_M4.89611.7 GBgood
Q5_K_S5.57696.7 GBgood
Q5_K_M5.7713.0 GBgood
Q6_K6.56820.5 GBexcellent
Q8_08.51063.0 GBlossless
FP16162000.5 GBlossless

Select your GPU above to see speed estimates and compatibility for each quantization.

▸ READY TO RUN THIS?RENT BY THE HOUR

RENT A GPU AND RUN KIMI K2.5 NOW

Rent on RunPod →Or Vast.ai →

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (13)

AIME96.1
τ²-Bench95.9
GPQA Diamond87.9
MMLU-PRO87.1
LiveCodeBench85.0
SWE-bench76.8
IFBench70.2
AA Long Context65.3
SciCode49.0
AA Intelligence46.8
AA Coding39.5
Terminal-Bench34.8
HLE30.1

Run this model

▸Easiest way to get started·Beginners
DOCS ↗
curl -fsSL https://ollama.com/install.sh | sh
$ollama run kimi-k2.5:q4_k_m

Downloads and runs automatically. Add --verbose for speed stats.

▸ SETUP GUIDE
>_

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more
Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines
HuggingFaceOllama LibraryGGUF DownloadsBuild Hardware

Find the best GPU for Kimi K2.5

Build Hardware for Kimi K2.5

Model Card

View on HuggingFace

Moonshot Kimi K2.5 — MoE model with strong reasoning and tool use capabilities.

▸ SPEC SHEET

Kimi K2.5 — 1000B MoE.

▸ SPECIFICATIONS
PARAMETERS
1000B (32B active)
ARCHITECTURE
Mixture of Experts
CONTEXT LENGTH
256K tokens
CAPABILITIES
chat, coding, reasoning, vision, tool_use
RELEASE DATE
2026-02-02
PROVIDER
Moonshot AI
FAMILY
kimi
▸ VRAM REQUIREMENTS
QUANTBPWVRAMQUALITY
IQ2_XXS2.38298.0 GB65%
IQ2_M2.93366.7 GB75%
Q2_K3.16395.5 GB78%
IQ3_XXS3.25406.7 GB82%
IQ3_XS3.5438.0 GB84%
Q3_K_S3.64455.5 GB85%
IQ3_M3.76470.5 GB86%
Q3_K_M4500.5 GB88%
Q3_K_L4.3538.0 GB90%
IQ4_XS4.46558.0 GB92%
Q4_K_S4.67584.2 GB93%
Q4_K_M4.89611.7 GB94%
Q5_K_S5.57696.7 GB96%
Q5_K_M5.7713.0 GB96%
Q6_K6.56820.5 GB97%
Q8_08.51063.0 GB100%
FP16162000.5 GB100%
§ 01BENCHMARK SCORES
MMLU-PRO87.1
GPQA Diamond87.9
LiveCodeBench85.0
AIME96.1
HLE30.1
AA Intelligence46.8
AA Coding39.5
SWE-bench76.8
aa_ifbench70.2
aa_terminal_bench34.8
aa_tau295.9
aa_scicode49.0
aa_lcr65.3
§ 02RUN COMMAND

Run Kimi K2.5 locally with Ollama — needs 611.7 GB VRAM at Q4_K_M:

$ollama run kimi-k2.5