MiniMax M2.7 230B-A10B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
230B total / 10B active across 256 experts. Needs roughly 130 GB at Q4 — a multi-GPU workstation or a rented node. The headline constraint is legal rather than technical: the weights are freely downloadable but the license is non-commercial, so it is not a drop-in for business use the way an Apache or MIT model would be. Context window is recorded as null here: published figures disagree (200K in most sources, 192K in the vLLM recipe) and neither could be confirmed against a primary page from this environment.
MiniMax M2.7 230B-A10B needs about 140 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
VRAM and speed by quantization
Quoted against NVIDIA RTX 4090 (24 GB). Weights plus overhead only: this model's architecture is unpublished, so no KV cache is added.
| Quant | Memory | VRAM | Speed (est.) | Fit |
|---|---|---|---|---|
| Q2_K 2.63 bpw | 76.4 GB | — | Too big | |
| Q3_K_M 3.41 bpw | 98.8 GB | — | Too big | |
| Q4_K_M 4.83 bpw | 139.7 GB | — | Too big | |
| Q5_K_M 5.67 bpw | 163.8 GB | — | Too big | |
| Q6_K 6.56 bpw | 189.4 GB | — | Too big | |
| Q8_0 8.50 bpw | 245.2 GB | — | Too big | |
| F16 16.00 bpw | 460.8 GB | — | Too big |
Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. MiniMax M2.7 230B-A10B VRAM calculator →
Get MiniMax M2.7 230B-A10B running
The cheapest catalogued GPU that runs MiniMax M2.7 230B-A10B is the Apple M3 Ultra (512 GB).
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
How to run MiniMax M2.7 230B-A10B
Install Ollama, then run:
ollama run minimax-m2-7Specifications
Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.
Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.
- Parameters
- 230 Billion (10B active)
- Context window
- Unverified — sources disagree (192K–200K)
- Architecture
- Mixture-of-Experts (256 experts, 62 layers, RoPE)
- Provider
- MiniMax
- Licence
- Non-Commercial (commercial use requires separate agreement)
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 256 GB
- Record updated
- 2026-08-15
Research / non-commercial only — this licence does NOT permit shipping a commercial product.
Quality and use cases
Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.
Can I run MiniMax M2.7 230B-A10B on my GPU?
- MiniMax M2.7 on AMD Ryzen AI Max+ 395
- MiniMax M2.7 on Apple M1 Max
- MiniMax M2.7 on Apple M1 Ultra
- MiniMax M2.7 on Apple M2 Max
- MiniMax M2.7 on Apple M2 Ultra
- MiniMax M2.7 on Apple M3 Max
- MiniMax M2.7 on Apple M4 Max
- MiniMax M2.7 on Apple M5 Max
- MiniMax M2.7 on Apple M5 Pro
- MiniMax M2.7 on NVIDIA A100 80GB (PCIe)
- MiniMax M2.7 on NVIDIA DGX Spark
- MiniMax M2.7 on NVIDIA H100 80GB (PCIe)
- MiniMax M2.7 on NVIDIA RTX PRO 6000 Blackwell
MiniMax M2.7 230B-A10B — frequently asked questions
How much VRAM does MiniMax M2.7 230B-A10B need?
About 140 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.
Does MiniMax M2.7 230B-A10B run on an RTX 4090 (24 GB)?
No. MiniMax M2.7 230B-A10B needs about 140 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.
How do I run MiniMax M2.7 230B-A10B locally?
Install Ollama and run `ollama run minimax-m2-7`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.