Kimi K2.5
Kimi K2.5 is a native multimodal MoE model created by continuing pretraining Kimi-K2-Base on approximately 15 trillion mixed visual and text tokens. A MoonViT vision encoder enables image and video understanding alongside coding and reasoning. Instant and thinking modes support both conversation and agent workloads. Moonshot also provides an Agent Swarm system that coordinates parallel agents; this is a harness capability around the model rather than additional model parameters. Public modified-MIT weights, an API, and Kimi chat access make the release available for research and deployment.
2026-01-27
1T total, 32B activated; 400M vision encoder
Mixture of Experts (MoE) with MoonViT vision encoder
Modified MIT
Specifications
- Parameters
- 1T total, 32B activated; 400M vision encoder
- Architecture
- Mixture of Experts (MoE) with MoonViT vision encoder
- License
- Modified MIT
- Context Window
- 262,144 tokens
- Type
- multimodal
- Modalities
- textimagevideo
Benchmark Scores
Advanced Specifications
- Model Family
- Kimi
- Finetuned From
- Kimi-K2-Base
- API Access
- Available
- Chat Interface
- Available
Capabilities & Limitations
- Capabilities
- reasoningcode generationtool usevisual reasoninginstant and thinking modesmultimodal coding
- Known Limitations
- The Agent Swarm experience depends on Moonshot orchestration and available tools.Custom commercial conditions apply to the weight license.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 developmentvisual interface development
- Function Calling Support
- Yes
- Tool Use Support
- Yes