Moonshot AI

Kimi K2.5

Name: Kimi K2.5
Author: Moonshot AI

Frontier

HuggingFace

1.4MDownloads2.8KLikesJan 2026Veröffentlicht256K TokenKontextCustomLizenz100 HerausragendQualität

Kimi K2.5 (1000B parameters) requires approximately 620.4 GB of VRAM with Q4_K_M quantization. As a Mixture of Experts model with 32B active parameters, it uses less memory than its total parameter count suggests. For the best balance of quality and speed, we recommend hardware with at least 714 GB of VRAM.

Loslegen

— kopieren & einfügen, um lokal auszuführen

Copy-paste commands to run Kimi K2.5 on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "moonshotai/Kimi-K2.5" \
  --hf-file "Kimi-K2.5-Q4_K_M.gguf" \
  -c 4096 -ngl 99

Quick specs

Parameters1000B (32B active)

Architecturemoe (MoE)

Context256K tokens

Modalitytext+vision

Min RAM390 GB

Rec. RAM610 GB (Q4_K_M)

LicenseCustom

FamilyKimi

✓ Vision✓ Code✓ Chat✓ Reasoning

About this model

Kimi K2.5 is Moonshot AI's advanced reasoning model with strong performance in math, coding, and multilingual tasks. Features long-context understanding and agentic capabilities for complex multi-step problem solving.

•Native Multimodality: Pre-trained on vision–language tokens, K2.5 excels in visual knowledge, cross-modal reasoning, and agentic tool use grounded...
•Coding with Vision: K2.5 generates code from visual specifications (UI designs, video workflows) and autonomously orchestrates tools for visual...
•Agent Swarm: K2.5 transitions from single-agent scaling to a self-directed, coordinated swarm-like execution scheme. It decomposes complex tasks...

Verwandte Modelle

Quantisierungsoptionen

VRAM-Schätzungen nach Quantisierungsstufe

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	390.0 GB	Low	—
Q3_K_S	3	490.0 GB	Low	—
NVFP4	4	560.0 GB	Medium	—
Q4_K_M	4	610.0 GB	Medium	—
Q5_K_M	5	720.0 GB	High	—
Q6_K	6	820.0 GB	High	—
Q8_0	8	1070.0 GB	Very High	—
F16	16	2050.0 GB	Maximum	—

Quality benchmarks

Kimi K2.5 benchmark scores

Benchmark verified

Coding

SWE-bench Verified76.8%

HumanEval+99.0%

Aider Polyglot—

LiveCodeBench53.7%

Reasoning

MMLU-Pro87.1%

GPQA Diamond87.6%

MATH-50098.0%

ARC Challenge—

General

Chatbot Arena—

IFEval94.0%

Source: official · 2026-01-01

Hardware-Kompatibilität

Eignungsschätzungen für alle Hardware

Rechner öffnen

Computing compatibility...

Speicheraufschlüsselung

Reference: RTX 2060 6GB

Weights610.0 GB

KV Cache7.4 GB

Runtime2.4 GB

Headroom0.6 GB

Häufig gestellte Fragen

FAQ — Kimi K2.5

Siehe auch

Quantisierungsleitfaden Bewertungsmethodik Rechner öffnen