// MiniMax · Vision-language
MiniMax-M3: VRAM & local deployment
428B total · 23B active (MoE) parameters. About 286 GB at 4-bit and 942 GB at BF16.
released 2026-06-02licence: Custom licence – read itmemory figures are estimates
01Specs
| Publisher | MiniMax · MiniMaxAI |
|---|---|
| Family / series | MiniMax · MiniMax-M3 |
| Type | Vision-language (image-text-to-text) |
| Parameters | 428B total · 23B active (MoE) Source: official model card“~428B parameters and ~23B activated parameters” |
| Context length | 1M tokens (1,048,576)max_position_embeddings in config.json |
| Architecture | 60 layers · GQA · MoE |
| Released weights | BF16 · 854 GBdownload |
| Licence | minimax-community · Custom licence – read itThe publisher uses its own licence; we have not classified it. Read the licence text. |
| Released | 2026-06-02 (Hugging Face repo created) |
02How much VRAM it needs (estimate)
Total = weights + 10% + 0.5GB overhead + KV cache for one request (FP16 KV). More concurrent users multiply the KV column.
| Precision | Weights | Smallest GPU | Mac |
|---|---|---|---|
| BF16 / FP16 (original) | 856 GB | 8 × H200 (1,128GB) | — |
| FP8 / INT8 / Q8_0 | 455 GB | 8 × H100 (640GB) | — |
| 4-bit (Q4_K_M / AWQ / GPTQ-Int4) | 259 GB | 4 × RTX PRO 6000 (384GB) | Mac 512GB |
KV cache not estimated: this model uses a new compressed / sparse attention design that our formula does not cover. Totals above are weights + overhead only.
Bits per weight: BF16 16, FP8 8.5, 4-bit 4.85. “Fits” = total ≤ 90% of memory; Mac GPU can use ~2/3 of unified memory (≤36GB) or ~3/4 (larger). For MoE, all experts must be in memory, so size by total parameters. Method →
03How to run it
Only runtimes the publisher or the runtime’s official library names for this model.
- vLLM — mentioned in the official model card
- SGLang — mentioned in the official model card
- Transformers — mentioned in the official model card
- KTransformers — mentioned in the official model card
04Official repositories & quantised builds
2 official repo(s) from MiniMaxAI.
- MiniMax-M3 main
- MiniMax-M3-MXFP8 main
05Other MiniMax models
All 11 →| Model | Params | 4-bit total | Released |
|---|---|---|---|
| MiniMax-Music3 | 2.43B | 2.1 GB | 2026-08-07 |
| MiniMax-H3 | 33.1B | 22.6 GB | 2026-07-28 |
| MiniMax-M2.7 | 228.7B | 155 GB | 2026-04-09 |
| MiniMax-M2.5 | 228.7B | 155 GB | 2026-02-12 |
| MiniMax-M2.1 | 228.7B | 155 GB | 2025-12-20 |
| MiniMax-M2 | 230B / 10B | 156 GB | 2025-10-22 |
| MiniMax-M1-80k | 456B / 45.9B | 305 GB | 2025-06-13 |
| MiniMax-M1-40k | 456B / 45.9B | 305 GB | 2025-06-05 |
06Sources
Data checked 2026-07-23. Parameters, context and licence are copied from the official repository; memory figures are our estimates.