Kimi K2 Instruct

moonshotai · released 2025-07-11 · other license

Kimi K2 Instruct is Moonshot's July 2025 1T-parameter (32B active) non-thinking open-weight model for chat and agentic tool use, with 65.8% SWE-bench Verified single-attempt and strong math/tool-use scores (vendor).

Key specs

TypeLocal open-weight
Parameters1026.47B total · MoE, — active
Architecturekimi_k2
Context window131K tokens
Knowledge cutoff2024-12-31
Modalitiestext
Recommended backends
Minimum viable rig

Benchmark scores

GPQA Diamond75.1%
SWE-bench Verified65.8%
AIME49.5%
MMLU-Pro81.1%
BFCL v3 (tool use)
Composite score6.89
Community ratingNo reviews yet

VRAM & disk per quantization

QuantVRAMDiskRAMContext
Q4_K_M596.9 GB595.4 GB131K

Strengths & weaknesses

Strengths: Strong non-thinking agentic and tool-use model; Excellent math (MATH-500 97.4) and instruction following (IFEval 89.8); 1T MoE, open weights

Weaknesses: No long chain-of-thought; AIME 2025 only 49.5; Weak terminal/agentic coding vs thinking models