LLMOpen weights
Kimi K2
Moonshot AI
Updated May 21, 2026 · Benchmarks via Artificial Analysis, specs & pricing via OpenRouter · Methodology · Report an error
Superseded by Kimi K2 Thinking
Specifications
- Type
- LLM
- Access
- Open weights
- Released
- July 11, 2025
- License
- Modified MIT
- Parameters
- 1T (32B active)
- Context window
- 131,072 tokens
- Knowledge cutoff
- December 31, 2024
- Output speed
- 26 tok/s
- Latency (TTFT)
- 1.51s
- Input
- Text
- Output
- Text
- API pricing
- $0.57 in · $2.30 out / 1M tokens
Capabilities
Function calling
Benchmarks
Reasoning
| GPQA Diamond | 76.6 |
Coding
| SWE-bench Verified | 65.8 |
| LiveCodeBench | 55.6 |
General
| MMLU | 89.5 |
| MMLU-Pro | 82.4 |
| Humanity’s Last Exam | 7 |
Our take
First trillion-parameter open-weights MoE, setting a new bar for open agentic and coding performance and introducing the MuonClip optimizer.