Kimi K2 Thinking
Kimi K2 Thinking is the reasoning specialization of the K2 trillion-parameter MoE model. It interleaves extended reasoning with tool calls and is trained to sustain long agent trajectories, with Moonshot demonstrating sequences of 200–300 tool calls. A 256K context supports research, programming, and mathematical tasks. Public weights use modified MIT terms and include native INT4 quantization. It is distinct from the original non-thinking K2-Instruct checkpoint; the reasoning and agent benchmark results belong to this model. Official API and Kimi access complement local inference.
2025-11-06
1T total, 32B activated
Mixture of Experts (MoE) with Multi-head Latent Attention
Modified MIT
Specifications
- Parameters
- 1T total, 32B activated
- Architecture
- Mixture of Experts (MoE) with Multi-head Latent Attention
- License
- Modified MIT
- Context Window
- 262,144 tokens
- Type
- text
- Modalities
- text
Benchmark Scores
Advanced Specifications
- Model Family
- Kimi
- API Access
- Available
- Chat Interface
- Available
Capabilities & Limitations
- Capabilities
- reasoningcode generationtool useinterleaved thinkingmathematics
- Known Limitations
- Long reasoning and agent trajectories can incur significant latency and token usage.The modified MIT license includes additional commercial conditions.Local deployment requires substantial memory for the full MoE weights.Agent performance depends on tools, harness configuration, and context management.
- Notable Use Cases
- coding assistantsmulti-step research agentssoftware development
- Function Calling Support
- Yes
- Tool Use Support
- Yes