Model

Kimi K2

OPEN

Moonshot AI · Kimi K2 family · released Jul 11, 2025

1T-parameter MoE agentic model trained with the MuonClip optimizer; open-sourced under a modified MIT license.

ReasoningCodingVisionFunction callingTool useAgentic

1526.6

Elo · rank #117

Parameters

1000B

Active params

32B (MoE)

Context

128K tokens

Architecture

Mixture-of-Experts, 384 experts (8 active + 1 shared/token), 61 layers, MLA attention

License

Modified MIT License

Languages

—

API price (in/out)

$0.57 / $2.3

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

Aider PolyglotCoding60.0%#19

best: Claude Opus 4.5 · 89.4%

AIMEMath69.6%#107

best: GPT-5.2 · 100.0%

Arena EloHuman preference1417#18

best: Claude Fable 5 · 1505

BFCL v3Agents25.3%#32

best: Hunyuan-A13B · 78.3%

GPQA DiamondReasoning75.1%#103

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning4.7%#122

best: Claude Sonnet 5 · 57.4%

IFEvalReasoning89.8%#17

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding53.7%#115

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math97.4%#15

best: GPT-5 · 99.4%

MMLU-ProKnowledge81.1%#52

best: Claude Fable 5 · 91.5%

MMLU-ReduxKnowledge92.7%#16

best: Qwen3.7-Max · 95.0%

MMLUKnowledge89.5%#15

best: OpenAI o3 · 92.9%

SimpleQAKnowledge31.0%#13

best: GPT-4.5 · 62.5%

SuperGPQAReasoning57.2%#16

best: Qwen3.7-Max · 73.6%

SWE-bench VerifiedCoding65.8%#68

best: Claude Fable 5 · 95.0%

τ²-Bench AirlineAgents56.5%#7

best: Trinity-Large-Thinking · 88.0%

τ²-Bench RetailAgents70.6%#10

best: Claude Sonnet 4.6 · 91.7%

τ²-Bench TelecomAgents65.8%#29

best: Claude Opus 4.6 · 99.3%

Terminal-BenchCoding25.0%#14

best: Claude Opus 4.5 (High) · 59.3%

Run it locally

VRAM @ Q4

600 GB

VRAM @ FP16

2000 GB

Fits on (Q4)

Multi-node cluster required

Does not fit on a single RTX 4090 (24GB); practical Q4 quants need ~250-600GB combined RAM/VRAM, with heavy CPU/SSD offload reported at only ~5-10 tok/s

Quantizations

GGUF

Kimi K2 family

Elo progression across releases

Kimi K2Jul 20251526.6 Kimi K2 ThinkingNov 20252177.2 Kimi K2.5Jan 20262253.5 Kimi K2.6Apr 20262517.3 Kimi K2.7 CodeJun 20262232.7

Resources

HFmoonshotai/Kimi-K2-Instruct GHmoonshotai/Kimi-K2 DOCSplatform.moonshot.ai APIAvailable via Moonshot AI API

API price $0.57/$2.3 · each benchmark row carries its own source badge (see methodology)