MiniMax M3 models
4 MiniMax M3 models from MiniMax on Hugging Face, from 3.1B to 440.3B parameters, published by MiniMax in BF16, with a 1M-token context. The smallest official model, MiniMax-M3, needs about 955 GB of VRAM at its published precision; the cheapest live fit is AMD MI300X at $13.14/hr.
Part of the MiniMax series · Previous generation: MiniMax M2
Pick a size
One row per official MiniMax M3 size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| MiniMax-M3 | 427.0B | 955 GB | 477 GB | 239 GB | AMD MI300X × 5 | $13.14/hr | 3.8 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| MiniMax-M3 | 427.0B | BF16 | 955 GB | AMD MI300X × 5 | $13.14/hr |
Quantized versions (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| MiniMax-M3-MXFP8 | 440.3B | F8_E4M3 | 492 GB | AMD MI300X × 3 | $7.89/hr |
Fine-tunes and community models (2)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| MiniMax-M3-EAGLE3-GQA | 3.1B | BF16 | 6.9 GB | RTX 3070 | $0.088/hr |
| MiniMax-M3-EAGLE3 | 3.3B | BF16 | 7.3 GB | RTX 3070 | $0.088/hr |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
Run and fine-tune MiniMax M3
- How to run MiniMax M3: engine support, chat template and context flags, with a serving command sized per model.
- How to fine-tune MiniMax M3: LoRA and QLoRA memory per model, and the license limits on derivatives.
More from MiniMax
- MiniMax H3 (8 models)
- MiniMax M2 (4 models)
- MiniMax Music3 (1 models)
- MiniMax-Text (1 models)