MiniMax-M3 Memory at each precision, the licence in plain words, and where it fits.
Short answer. MiniMax-M3 needs about 243 GB of memory at 4-bit, 456 GB at 8-bit and 857 GB at full 16-bit precision, including runtime overhead and an 8,192-token KV cache. It has about 427 billion parameters and a 1,048,576-token context window. Licence in short: custom licence: check before commercial use. By memory alone, on ANVE’s launch configurations: ANVE One does not fit, ANVE Office does not fit, ANVE Office Pro (GPU) does not fit. These are estimates from its published size, not benchmarks.
01
What it is
MiniMax-M3 is a vision-language model: it reads images (scanned pages, photos, charts and screenshots) as well as text, published by MiniMax on Hugging Face on 2 June 2026. It has about 427 billion parameters and a configured context window of 1,048,576 tokens. It is a mixture-of-experts model: 128 experts, 4 used for each token.
Its card lists the licence as a custom licence (minimax-community). In short, the licence sets its own terms; read it before any commercial use.
Hugging Face counted about 200,665 downloads of it in the 30 days before we checked (7 Oct 2026), and 1,565 people have liked it there.
02
How much memory it needs
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 0.12 MB of KV cache at 16-bit, so 32,768 tokens would add about 4.0 GB.
Precision
Weights
KV cache (8,192 tokens)
Total, estimated
BF16 (full precision)
854 GB
1.0 GB
857 GB
8-bit (Q8_0)
454 GB
1.0 GB
456 GB
4-bit (Q4_K)
240 GB
1.0 GB
243 GB
03
Architecture at a glance
Setting
Value in config.json
Architecture class
MiniMaxM3SparseForConditionalGeneration
Model type
minimax_m3_vl
Layers
60
Hidden size
6,144
Attention heads
64
Key-value heads
4 (grouped-query attention)
Head size
128
Vocabulary
200,064
Experts
128, 4 active per token
Configured context
1,048,576
04
Will it fit on ANVE?
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.
Configuration (at launch)
Memory for the model
Verdict
ANVE One
about 48 GB of 64 GB unified memory
does not fit
ANVE Office
24 GB on the RTX PRO 4000 Blackwell SFF Edition
does not fit
ANVE Office Pro (GPU)
24 GB on the RTX PRO 4000 Blackwell
does not fit
05
What an office would use it for
This is frontier-scale: even at 4-bit it needs far more memory than any single office box, so it belongs on a multi-GPU server or a cloud service, not on an appliance.
06
Similar models to compare
Qwen3.5-397B-A17B (403.4B parameters, about 228 GB at 4-bit before context; commercial use allowed)
GLM-5.3-Flash (321.3B parameters, about 182 GB at 4-bit before context; commercial use allowed)
DeepSeek-V4.1-Flash (763.2B parameters, about 431 GB at 4-bit before context; commercial use allowed)
07
Questions people ask
How much memory does MiniMax-M3 need?
About 243 GB at 4-bit, 456 GB at 8-bit and 857 GB at full 16-bit precision, including an 8,192-token KV cache and about 1.5 GB of runtime overhead. These are estimates from its parameter count.
Can MiniMax-M3 run on a 24 GB graphics card?
Not entirely: even at 4-bit it needs about 243 GB, so part of it would sit in system memory, which is much slower.
Is MiniMax-M3 free for commercial use?
Its card lists a custom licence (minimax-community), which sets its own terms; read it before any commercial use. Read the licence itself before relying on this summary; it is not legal advice.
What is MiniMax-M3’s context window?
Its configuration sets 1,048,576 tokens. Longer contexts need more memory for the KV cache, so the practical limit on a given machine can be lower.
08
What is not real yet
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.
