Kimi-VL
Kimi-VL is a compact vision-language MoE family using the native-resolution MoonViT visual encoder and a Moonlight language backbone. It supports image and video comprehension, OCR, multi-image questions, visual reasoning, and agent tasks. The language decoder activates roughly 2.8B parameters. Instruct and Thinking releases are grouped here, including the June 2025 Thinking update. Public MIT weights and inference code support local experimentation; long-thinking variants trade additional generation for more difficult reasoning.
2025-04-10
16B total; approximately 2.8B language parameters activated
MoonViT vision encoder, MLP projector, and Mixture of Experts language decoder
MIT
Specifications
- Parameters
- 16B total; approximately 2.8B language parameters activated
- Architecture
- MoonViT vision encoder, MLP projector, and Mixture of Experts language decoder
- License
- MIT
- Context Window
- 131,072 tokens
- Type
- multimodal
- Modalities
- textimagevideo
Benchmark Scores
Advanced Specifications
- Model Family
- Kimi
- Finetuned From
- Moonlight-16B-A3B
- API Access
- Not Available
- Chat Interface
- Not Available
- Variants
- Kimi-VL-A3B-InstructKimi-VL-A3B-ThinkingKimi-VL-A3B-Thinking-2506 (2025-06-21)
Capabilities & Limitations
- Capabilities
- visual reasoningOCRmulti-image understandingvisual agents
- Known Limitations
- Thinking checkpoints can produce long reasoning traces.The 2.8B activated figure refers to the language decoder, not the complete model.
- Notable Use Cases
- document question answeringvisual reasoning researchimage-grounded assistants