Kimi-VL-A3B-Thinking
Moonshot AI · 2025-04-10 · 16.4B parameters
Kimi-VL-A3B-Thinking is Moonshot AI's open-weight long-thinking Mixture-of-Experts vision-language checkpoint, released with the Kimi-VL technical report in April 2025. It adds long chain-of-thought supervised fine-tuning and reinforcement learning for multimodal reasoning atop a native-resolution MoonViT encoder, MLP projector, and MoE language decoder; the official report describes 2.8B activated language-model parameters plus a 400M vision encoder and a 131,072-token extended context window.
Benchmark scores
| Benchmark | Score |
|---|---|
| MATH-Vision | 36.8 |
| OSWorld-Verified | 10.3 |