Loading...
Loading...
Kimi K2 0905 is the September update of [Kimi K2 0711](moonshotai/kimi-k2). It is a large-scale Mixture-of-Experts (MoE) language model developed by Moonshot AI, featuring 1 trillion total parameters with 32 billion active per forward pass. It supports long-context inference up to 256k tokens, extended from the previous 128k. This update improves agentic coding with higher accuracy and better generalization across scaffolds, and enhances frontend coding with more aesthetic and functional outputs for web, 3D, and related tasks. Kimi K2 is optimized for agentic capabilities, including advanced tool use, reasoning, and code synthesis. It excels across coding (LiveCodeBench, SWE-bench), reasoning (ZebraLogic, GPQA), and tool-use (Tau2, AceBench) benchmarks. The model is trained with a novel stack incorporating the MuonClip optimizer for stable large-scale MoE training.
Quality Index
30.9
95th of 442
Top 21%
Coding Index
25.9
99th of 352
Top 28%
Math Index
57.3
122nd of 268
Top 46%
Price/1M
$1.14
498th cheapest
267% above median
Top 73%
Speed
54 tok/s
Top 45%
TTFT
0.78s
Context Window
131K
145th largest
Top 63%
Input
$0.80
per 1M tokens
Output
$2.25
per 1M tokens
Blended
$1.14
per 1M tokens
Cheaper than 27% of models. Median price is $0.31/1M tokens.
Daily
$1.14
Monthly
$34.11
54
tokens/sec
Faster than 55% of models
0.78
seconds
Faster than 36% of models
0.78
seconds
Faster than 41% of models
Market Median
46 tok/s
19% faster
Median TTFT
0.42s
87% slower
Throughput/Dollar
48
tok/s per $/1M
Speed Comparison
Context Window
131K
tokens
Larger than 37% of models