Kimi-VL
16B mixture-of-experts vision-language model with a 128K context window
About
Kimi-VL is Moonshot AI's open vision-language model from April 2025: 16B total parameters with about 3B activated per token, a native-resolution MoonViT visual encoder, and 128K context. It covers general perception, OCR, long video, long documents, and OS-agent grounding, with a separate Thinking variant for multimodal reasoning. The team reports 83.2 on the V* high-resolution perception benchmark and 52.8 on ScreenSpot-Pro. It is released under the plain MIT license, which is unusually permissive for a model this capable. The ecosystem is thinner than Qwen's, and long-video inference is memory-hungry despite the small active size.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
V* Benchmark 83.2, ScreenSpot-Pro 52.8, OSWorld-G 52.5 (Moonshot-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: April 2025 release, 16B/3B-active MoE, 128K context, MIT license verified across Moonshot repo mirror and model timeline sources; V* Benchmark 83.2 and ScreenSpot-Pro 52.8 are Moonshot-reported; labeled as such; Thinking variant (June 2025) noted; ecosystem thinner than Qwen's.