Kimi K2.6 is Moonshot AI's open-weight flagship model. It is a 1T-parameter Mixture-of-Experts architecture with 32B active parameters per token, a 256K context window, and native INT4 quantization via Quantization-Aware Training, offering improved inference speed and reduced GPU memory usage. It accepts text and image input and is tuned for long-horizon agentic coding, supporting continuous work, sequential tool calls without drift, and large-scale agent swarms. It is designed for coding, motion-rich frontend generation, and complex multi-agent workflows.
VisionReasoning|Proprietary Model
Knowledge Cutoff
Unknown
Input → Output Format
Context Memory
256KIN66KOUT
AI Performance Evaluation
Arena Overall Score
1460
±7As of 2026-05-01
Overall Rank
No.28
6,168 Votes
Arena by Ability
Hard Prompts
1485±9No.26
Expert Knowledge
1493±23No.20
Instruction Following
1455±13No.27
Conversation Memory
1460±18No.42
Creative
1424±20No.41
Coding
1516±15No.20
Math
1468±28No.24
Arena by Occupation
Creative Writing
1437±16No.37
Social Sciences
1473±17No.32
Media
1430±18No.35
Business
1449±16No.37
Healthcare
1480±26No.29
Legal
1435±26No.72
Software
1497±12No.26
Mathematics
1463±29No.31
Source:Arena Intelligence
Overall
AA Intelligence Index
54%↑15%
LiveBench
72%↑12%
Reasoning & Math
GPQA Diamond
91%↑9%
HLE
36%↑18%
LB Reasoning
79%↑10%
LB Math
84%↑10%
LB Data
65%↑12%
Coding
AA Coding Index
47%↑11%
LB Coding
79%↑6%
LB Agentic
58%↑13%
TAU2
96%↑15%
TerminalBench
44%↑10%
SciCode
54%↑12%
Language & Instructions
IFBench
76%↑13%
AA-LCR
70%↑8%
LB Language
75%↑3%
LB IF
64%↑13%
Output Speed
Standard Mode
37tok/s↓41
First Output 1.48s
