Kimi-VL-A3B-Instruct
Memory at each precision, the licence in plain words, and where it fits.

Kimi-VL-A3B-Instruct is a vision-language model: it reads images (scanned pages, photos, charts and screenshots) as well as text, published by Moonshot AI on Hugging Face on 9 April 2025. It has about 16.4 billion parameters and a configured context window of 131,072 tokens. It is a mixture-of-experts model: 64 experts, 6 used for each token, so only about 3B parameters are active at a time.

Its card lists the licence as MIT. In short, the licence allows commercial use, modification and redistribution as long as the copyright notice is kept.

Hugging Face counted about 232,154 downloads of it in the 30 days before we checked (7 Oct 2026), and 284 people have liked it there.

It is built on moonshotai/Moonlight-16B-A3B, according to its card.

We size it the way you would size any model: the weights (parameters × bits ÷ 8), the KV cache for 8,192 tokens of context, and roughly 1.5 GB for the runtime itself. Real use varies with settings.

Each token of context adds about 0.03 MB of KV cache at 16-bit, so 32,768 tokens would add about 1.0 GB. This model uses multi-head latent attention, which stores a compressed cache.

These verdicts are arithmetic, not tests, against the specification at launch: ANVE One can give a model about three quarters of its 64 GB of unified memory, while ANVE Office and ANVE Office Pro hold it on a 24 GB graphics card and keep the rest of the stack in 64 GB or 128 GB of system memory. ANVE’s tested default is Qwen 3.8 27B; this model is not yet validated on ANVE.

This is the class ANVE runs by default (27B-class): strong enough for meeting minutes, summaries, drafting and question answering over an office’s own documents, and small enough for one box.

• Apriel-1.6-15b-Thinker (14.9B parameters, about 9.9 GB at 4-bit before context; commercial use allowed)
• gemma-3-12b-it (12.2B parameters, about 8.4 GB at 4-bit before context; commercial use allowed, with conditions)
• Llama-3.2-11B-Vision-Instruct (10.7B parameters, about 7.5 GB at 4-bit before context; commercial use allowed, with conditions)

No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.

Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.