Models / MiniMax M2.7
Can I run MiniMax M2.7 locally?
MiniMax · 229B parameters · mixture of experts · up to 200K context · license other · Hugging Face
At Q4_K_M with a 32K context, MiniMax M2.7 needs about 153 GB of GPU or unified memory by our estimate. The official checkpoint (FP8) is 230 GB on disk.
Memory by quantization and context
| Quantization | Weights | Total at 8K | Total at 32K | Total at 128K |
|---|---|---|---|---|
| Q3_K_M (~4.0 bit) | 114 GB | 121 GB | 127 GB | 152 GB |
| Q4_K_M (~4.9 bit, popular default) | 140 GB | 147 GB | 153 GB | 178 GB |
| Q5_K_M (~5.7 bit) | 163 GB | 171 GB | 177 GB | 202 GB |
| Q6_K (~6.6 bit) | 188 GB | 196 GB | 203 GB | 228 GB |
| Q8_0 (~8.5 bit, near-lossless) | 243 GB | 253 GB | 260 GB | 285 GB |
| F16 / BF16 (16 bit) | 457 GB | 474 GB | 480 GB | 505 GB |
Bits per weight from llama.cpp's quantize docs. F16 KV cache. Totals include our runtime-overhead estimate.
Cheapest hardware that fits (Q4_K_M, 32K)
- Apple Mac Studio M3 Ultra 256GB (2025): about €6,724 (used, apple.com, 192 GB usable)
- Apple Mac Studio M5 Ultra 256GB (2026): about €9,830 (new, abt.com, 192 GB usable)
- 4× NVIDIA A100 40GB PCIe: about €14,191 (new, ebay.com.au, 156 GB usable)
- 4× NVIDIA A40 48GB: about €17,019 (used, ebay.co.uk, 188 GB usable)
- 4× NVIDIA RTX A6000 48GB: about €17,397 (used, ebay.ca, 188 GB usable)
Runs in
| Tool | Platforms | How | Source |
|---|---|---|---|
| LM Studio Desktop app | macOS, Windows (x64, arm64), Linux (x64, arm64) | lmstudio-community build: lmstudio-community/MiniMax-M2.7-GGUF | source |
| llama.cpp Inference engine (C/C++) | Linux, Windows, macOS, Android (build) | GGUF: bartowski/MiniMaxAI_MiniMax-M2.7-GGUF | source |
| MLX (mlx-lm) Apple framework + CLI | macOS (Apple Silicon only) | MLX conversion: mlx-community/MiniMax-M2.7-4bit | source |
| vLLM Serving engine | Linux | named in the official model card | source |
| SGLang Serving engine | Linux | named in the official model card | source |
Ollama: cloud only, it does not download the model to your machine.
Only tools with a public source are listed: an official build (ggml-org, lmstudio-community, mlx-community, bartowski or Unsloth on Hugging Face), the model card, ollama.com or the tool's own README. Apps built on llama.cpp, such as Jan and PocketPal AI, can usually load the same GGUF files. All local apps.
Try other settings
5 hardware options fit
Cheapest real offer per device from AI Hardware Deals, converted to EUR at approximate rates. Multi-GPU rows multiply the card price.
| Setup | Headroom | ||||
|---|---|---|---|---|---|
| Apple Mac Studio M3 Ultra 256GB (2025) unified memory | 1 device | 192 GB | 38.6 GB | €6,724 used, apple.com | n/a |
| Apple Mac Studio M5 Ultra 256GB (2026) unified memory | 1 device | 192 GB | 38.6 GB | €9,830 new, abt.com | n/a |
| NVIDIA A100 80GB PCIe server GPU (passive cooling) | 2× cards | 158 GB | 4.6 GB | €29,998 used, ebay.nl | n/a |
| NVIDIA DGX Station (GB300) unified memory | 1 device | 740 GB | 587 GB | €83,738 new, configurator.exxactcorp.com | n/a |
| Apple Mac Studio M5 Ultra 512GB (2026) unified memory | 1 device | 384 GB | 231 GB | no verified price | n/a |
Speed ceiling is memory bandwidth divided by the bytes of (active) weights read per token, shown only where we have a sourced bandwidth figure. Real decode speed is lower, especially with long contexts or when splitting across cards. All memory numbers are estimates; see the methodology.
FAQ
- How much VRAM does MiniMax M2.7 need?
- About 153 GB at Q4_K_M with a 32K-token context (140 GB weights, 8.3 GB KV cache, 5.2 GB overhead). These are estimates.
- Can I run MiniMax M2.7 on a single GPU?
- Not at Q4_K_M with 32K context on any single GPU in our catalog. Use several GPUs, a high-memory unified-memory machine, a smaller quantization or a shorter context.
- Which apps can run MiniMax M2.7 locally?
- Verified so far: LM Studio, llama.cpp, MLX (mlx-lm), vLLM, SGLang. Each entry links to its source on this page.