MiniMax M2.5 230B — VRAM, speed & local setup
Written by Jakub Rusinowski · Last updated
230B MoE model with 80.2% SWE-bench score — surpassing Claude Opus 4.6 on that benchmark at a fraction of the cost ($0.30/$0.56 per 1M tokens vs Claude's $15/$75). Supports 1M token context and multimodal input. Needs ~130 GB at Q4 — accessible via Mac Studio 192 GB or 2-4× A100 cluster.
MiniMax M2.5 230B 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 Weights plus overhead only: this model's architecture is unpublished, so no KV cache is added.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 | 139.6 GB | — | Too big | |
| Q5_K_M 5.67 bpw | 163.7 GB | — | Too big | |
| Q6_K 6.56 bpw | 189.3 GB | — | Too big | |
| Q8_0 8.50 bpw | 245.1 GB | — | Too big | |
| F16 16.00 bpw | 460.6 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 M2.5 230B VRAM calculator →
Get MiniMax M2.5 230B running
The cheapest catalogued GPU that runs MiniMax M2.5 230B 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 M2.5 230B
Install Ollama, then run:
ollama run minimax-m2-5Specifications
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
- 229.9B (9.8B active)
- Context window
- 1,000,000
- Architecture
- Mixture-of-Experts (256 experts, top-8, 62 layers)
- Provider
- MiniMax
- Licence
- Modified MIT (attribution required)
- Practical deployment
- Workstation (large unified memory or pro GPU)
- Specified at
- Q4_K_M
- System RAM
- 256 GB
- Record updated
- 2026-02-15
Commercial use permitted. No usage restrictions beyond attribution.
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 M2.5 230B on my GPU?
- MiniMax M2.5 on AMD Ryzen AI Max+ 395
- MiniMax M2.5 on Apple M1 Max
- MiniMax M2.5 on Apple M1 Ultra
- MiniMax M2.5 on Apple M2 Max
- MiniMax M2.5 on Apple M2 Ultra
- MiniMax M2.5 on Apple M3 Max
- MiniMax M2.5 on Apple M4 Max
- MiniMax M2.5 on Apple M5 Max
- MiniMax M2.5 on Apple M5 Pro
- MiniMax M2.5 on NVIDIA A100 80GB (PCIe)
- MiniMax M2.5 on NVIDIA DGX Spark
- MiniMax M2.5 on NVIDIA H100 80GB (PCIe)
- MiniMax M2.5 on NVIDIA RTX PRO 6000 Blackwell
MiniMax M2.5 230B — frequently asked questions
How much VRAM does MiniMax M2.5 230B 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 M2.5 230B run on an RTX 4090 (24 GB)?
No. MiniMax M2.5 230B 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 M2.5 230B locally?
Install Ollama and run `ollama run minimax-m2-5`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.