Can you run MiniMax M3 locally?
MiniMax family · 2026 · 427B parameters (23B activated per token) · Hugging Face model card
MiniMax M3 is a mixture-of-experts vision-language model with about 23B activated parameters per token and 427B total open weights (vendor card: 428B parameters and ~23B activated; HF safetensors.total=427,040,140,160 - the small difference is the announced-rounding, not a different model). Config reports 128 local experts with 4 experts per token plus 1 shared expert. Native context is 1,048,576 tokens. License is other (see repository). Local VRAM fit uses the 23B active count, not the 427B dense-equivalent: roughly 16 GB at Q4-class (calculated), a high-memory workstation. Weights still occupy ~428B-class disk. Measured speeds: not yet published here.
Minimum: 16 GB+ memory (Q4-class on 23B active, calculated) · Recommended: 24 GB+ for comfortable context headroom (Q4-class on 23B active, calculated)
How much VRAM does MiniMax M3 need at each quantization?
MiniMax M3 needs 282.3 GB of VRAM at Q4_K_M. The table below lists weights-only size and total VRAM including overhead for each common quantization level with 4k and 32k token contexts.
| Quantization | Bits / weight | Weights only | Total + KV @4k ctx | Total + KV @32k ctx |
|---|---|---|---|---|
| Q4_K_M | 4.8 | 256.2 GB | 282.3 GB | 285.9 GB |
| Q5_K_M | 5.7 | 304.3 GB | 335.2 GB | 338.7 GB |
| Q6_K | 6.6 | 352.3 GB | 388.0 GB | 391.6 GB |
| Q8_0 | 8.5 | 453.7 GB | 499.6 GB | 503.1 GB |
| FP16 | 16 | 854.1 GB | 940.0 GB | 943.5 GB |
GQA 64 heads / 4 KV heads, head_dim 128 (config)
Method: weights = parameters × bits-per-weight (Q4_K_M ≈ 4.8, Q5_K_M ≈ 5.7, Q6_K ≈ 6.6, Q8_0 ≈ 8.5, FP16 = 16), plus 10% loading overhead, plus KV cache from the verified architecture config (60 layers, 4 KV heads, 128 head dim). Source: HF-verified 2026-09-27 via api (gated=false private=false params=427.04B safetensors.total=427040140160 downloadable_safetensors_sum=854176398808 ctx=1048576 license=other repo=MiniMaxAI/MiniMax-M3@f0e1c1e04d40177e4673a22097036854f536e9c0 layers=60 kv_heads=4). MoE: num_experts=128 num_experts_per_tok=4 active_params_b=23.0 (vendor card; VRAM fit uses active, not dense-equivalent). VRAM figures are documented calculations, not measurements. No tok/s invented.
Which GPUs can run MiniMax M3 locally?
At Q4_K_M with a 4k context, MiniMax M3 needs 282.3 GB of VRAM. The lists below are computed live from our GPU database and grouped by how much headroom the card has. Disclosure: CompareAIHardware.com participates in the Amazon Associates program and earns from qualifying purchases through links on this page. Affiliate relationships do not influence our recommendations.
Minimum GPUs that fit MiniMax M3 at Q4_K_M
These GPUs hold the model but leave little headroom — keep contexts short.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| AMD Instinct MI355X | 288 GB | 8000 GB/s | Data Center GPU | |
| NVIDIA B300 (Blackwell Ultra) | 288 GB | 8000 GB/s | Data Center GPU |
Needs 2+ GPUs to run MiniMax M3
One of these cards is too small on its own, but a pair (tensor or pipeline parallel, ~90% efficiency) covers the 282.3 GB requirement. See our multi-GPU guide for setup.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| AMD Instinct MI300X | 192 GB ×2 | 5300 GB/s | Data Center GPU | |
| NVIDIA B200 | 192 GB ×2 | 8000 GB/s | Data Center GPU | |
| AMD Instinct MI325X | 256 GB ×2 | 6000 GB/s | Data Center GPU |
Can a Mac run MiniMax M3?
Yes — these Apple Silicon machines fit MiniMax M3 at Q4_K_M, since macOS lets the GPU use about 75% of unified memory. Generation speed is bound by memory bandwidth, so the GB/s column matters as much as capacity.
| Apple Silicon | Unified memory | Usable by GPU (~75%) | Bandwidth |
|---|---|---|---|
| M5 Ultra (Mac Studio) | 512 GB | 384 GB | 1200 GB/s |