Vision-language · Moonshot AI

Kimi-VL A3B

Hugging Face: moonshotai/Kimi-VL-A3B-Instruct · checked 2026-10-13

16B MoE vision-language model with 3B active parameters and 128K context.

Datasheet

Maker	Moonshot AI
Parameters	16.41 BVerifiedsafetensors.total from the HF API Source
Licence	MITVerified
Access	Public download
Task (HF pipeline)	image-text-to-text
First published on HF	2025-04-09
Last updated on HF	2026-01-30
Weights in memory	33 GB at 16-bit · 9.9 GB at ~4.85 bitsEstimateParameters × bits ÷ 8. Excludes context cache and runtime overhead; not every model has a 4-bit build.

Benchmark

57%

MMMU (val) · #2 of 3 scored in this category

MMMU-Val pass@1, model card table. Source

See the full vision-language ranking

The description summarises the model card at https://huggingface.co/moonshotai/Kimi-VL-A3B-Instruct. Parameters and licence are read from the Hugging Face API; the weights-in-memory line is our arithmetic, not a measurement.