Written by Jakub Rusinowski · Last updated August 15, 2026
MiniMax's March 2026 agentic MoE: 230B total parameters with 10B active per token across 256 experts — a 4.3% activation rate — over 62 layers, using multi-head causal self-attention with RoPE. Weights are published openly on Hugging Face and GitHub, so this is self-hostable rather than API-only, though the scale still implies a multi-GPU box. NVIDIA has upstreamed fused QK RMS Norm and FP8 MoE kernels from TensorRT-LLM for it. ⚠️ Released under a NON-COMMERCIAL license — commercial deployment requires a separate agreement with MiniMax.
| Licence | What it permits | Applies to |
|---|---|---|
CC-BY-NC-4.0 | Research / non-commercial only Research / non-commercial only — this licence does NOT permit shipping a commercial product. | MiniMax M2.7 230B-A10B |
| MiniMax M2.7 230B-A10B | Min 140 GB VRAM · Q4_K_M · Unverified — sources disagree (192K–200K) ctx · |
The cheapest GPU that runs MiniMax M2.7 locally (min 140 GB VRAM) is the Apple M2 Ultra (192 GB).
Install Ollama then run: ollama run
Minimum VRAM: 140 GB. For best results use Q4_K_M quantization.
MiniMax M2.7 needs about 140 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: MiniMax M2.7 230B-A10B (140 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
MiniMax M2.7's smallest variant needs about 140 GB, which exceeds a single RTX 4090 (24 GB). Use multiple GPUs, a higher-VRAM card, or Apple Silicon with large unified memory.
Q4_K_M is the best balance of quality and VRAM for MiniMax M2.7 in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
Install Ollama, then run: ollama run . This downloads MiniMax M2.7 and starts a local, OpenAI-compatible endpoint — no internet connection is needed after the initial download.