EDGE OSMINIMAX · 229B PARAMETERS
MiniMax-M2.5 hardware and VRAM requirements
MiniMax’s 229B mixture-of-experts model for coding and agent work, with over a million downloads last month. MiniMax-M2.5 is a mixture-of-experts model with only part of its 229B parameters works on each token. At FP8 it needs about 244 GB of GPU memory for one user, and about 275 GB to serve a team of 16 with 8K context each.
Total parameters
229B
Active per token
Not stated
Context length
Not stated on the model card
Released as
FP8
Minimum VRAM
135 GB at 4-bit
For a team of 16
275 GB at FP8
VRAM needed to run MiniMax-M2.5
GPU memory by precision, number of concurrent users and context length. Each user needs their own KV cache, so memory grows with users and context.
| Precision | Weights only | 1 user, 8K | 16 users, 8K | 16 users, 32K | LoRA fine-tune |
|---|---|---|---|---|---|
| FP8 Released | 229 GB | 244 GB | 275 GB | 375 GB | 296 GB |
| BF16 Upcast | 458 GB | 484 GB | 515 GB | 615 GB | 537 GB |
| 4-bit Community quantization (estimated) | 126 GB | 135 GB | 167 GB | 266 GB | 188 GB |
Full fine-tuning in mixed precision needs about 3.71 TB: weights, gradients and optimizer states at roughly 16 bytes per parameter. LoRA trains small adapter layers on a frozen copy of the model instead, so it needs far less.
GPUs that can run MiniMax-M2.5 on-premise
For a team of 16 using MiniMax-M2.5 at FP8 with 8K context each (about 275 GB), smallest fit first.
NVIDIA DGX B200
1.44 TB per system · 1.02 TB to spare.
NVIDIA H100 NVL
94 GB per GPU · Split across 4 GPUs in one server, 64 GB to spare.
NVIDIA L40S
48 GB per GPU · Split across 8 GPUs in one server, 71 GB to spare.
NVIDIA RTX PRO 6000 Blackwell
96 GB per GPU · Split across 4 GPUs in one server, 71 GB to spare.
AMD Instinct MI300X
192 GB per GPU · Split across 2 GPUs in one server, 71 GB to spare.
AMD Instinct MI325X
256 GB per GPU · Split across 2 GPUs in one server, 186 GB to spare.
NVIDIA H200 NVL
141 GB per GPU · Split across 4 GPUs in one server, 233 GB to spare.
NVIDIA DGX Spark
128 GB per system · Needs more than 2 linked systems.
Some of these setups need several GPUs working as one. Buying the GPUs is the easy part. Vanik Edge OS splits the model across them, keeps it running if a server fails and manages it all from one console. See how Edge OS clusters GPU servers
Weighing two of these? Compare AI hardware side by side, for example H100 vs H200, RTX PRO 6000 vs RTX 5090 or DGX Spark vs Mac Studio.
Running MiniMax-M2.5 for your team?
We will size the GPU server for your users and data, and set up MiniMax-M2.5 on it inside your own building. Free, no obligation.
MiniMax-M2.5 at a glance
Compare MiniMax-M2.5 with other models
FROM SIZING TO RUNNING
Sizing the hardware is the easy part.
Getting several GPUs to work as one system for a whole company is where most projects slow down. That is the part Vanik Edge OS does for you.
STEP 01 · THIS TOOL
Size it
Work out the GPU memory your models need and which GPUs or servers can hold them.
STEP 02 · VANIK EDGE OS
Cluster it
Edge OS joins your GPU servers into one cluster, splits a large model across them and keeps it running if a server fails. NVIDIA and AMD servers work side by side.
STEP 03 · VANIK EDGE OS
Run it
Your team gets AI apps with single sign-on and an audit trail, managed from one console, with your data staying inside your building.
MiniMax-M2.5 GPU questions
How much VRAM to run MiniMax M2.5?
MiniMax-M2.5 needs about 244 GB of GPU memory at FP8 for one user with 8K context: 229 GB for the weights plus the KV cache and runtime overhead. Serving 16 people at once with 8K context each needs about 275 GB.
Can I run MiniMax M2.5 locally?
Yes, on data-center GPUs. At 4-bit it needs about 135 GB of GPU memory for one user, more than consumer graphics cards or laptops hold. A team of 16 fits on NVIDIA DGX B200.
How big is MiniMax M2.5?
MiniMax-M2.5 has 229B parameters in total, in a mixture-of-experts design. The weights take about 126 GB at 4-bit.
Is MiniMax M2.5 open source?
Yes. MiniMax publishes the MiniMax-M2.5 weights on Hugging Face under a modified MIT license, so you can download and self-host it. Check the license on the model page for the exact terms.
How many GPUs does MiniMax M2.5 need?
For 16 users with 8K context at FP8 it needs about 275 GB. That fits on NVIDIA DGX B200.
Can I fine-tune MiniMax M2.5?
LoRA fine-tuning at FP8 needs roughly 296 GB with 4K sequences; it trains small adapter layers on a frozen copy of the model. Full fine-tuning needs about 3.71 TB, which calls for a multi-GPU training cluster.
How are these numbers calculated?
Weights are parameters × bytes per parameter. The KV cache uses 248.0 KB per token per user. We add about 5% of the weights plus 1 GB for runtime overhead and keep 10% of each GPU free. Figures are planning estimates; real usage depends on the inference engine and settings.
VRAM for similar models
Specifications from Hugging Face: MiniMaxAI/MiniMax-M2.5. Memory and bandwidth are manufacturer figures; GPU counts are rounded to 1, 2, 4 or 8 per server. All figures are planning estimates. Last checked November 10, 2026.
Run MiniMax-M2.5 inside your own building
Vanik Edge OS runs open-weight models on NVIDIA DGX, AMD Instinct and other GPU servers on your premises, so your data never leaves.