Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

MiniMax M2.5 GPU requirements: VRAM and cheapest GPU

MiniMax M2.5 has 228.7B parameters, published in FP8. It is a mixture-of-experts model: 8 of its 256 experts work on each token, about 11B parameters, so it generates at roughly the speed of a model that size. Every expert still has to sit in GPU memory, which is why the memory figures below are for the full model. Its licence (modified-mit) adds terms of its own, so read it before you use the model commercially.

Parameters
228.7B
Smallest footprint (INT4)
about 135 GB
Cheapest GPU per hour (FP16)
$2.39
Type
Text
  • Parameters 228.7B
  • Active per token 11B
  • Type Text
  • Task text-generation
  • Max context 192K
  • Licence modified-mit
  • Downloads 907.6K
  • Likes 1.5K

Calculating for MiniMax M2.5

Precision
Workload
More settings
Serving engine
Requests at once

MiniMax M2.5 needs about 506 GB of GPU memory

Inference (vLLM), FP16 / BF16, 8K tokens context, one request at a time.

Where the memory goes

  • Weights 457 GB
  • KV cache 2.1 GB
  • Runtime overhead 47 GB
  • Total 506 GB

GPU counts keep 10% of each card free for spikes. Plan extra if you expect long prompts or many users at once.

GPUs that fit, cheapest firstEvery GPU type we track that can hold this setup on one node, ranked by total hourly cost
RankConfigurationPer GPU / hrGPUs neededTotal / hrTotal / month (24/7)Provider
13 × MI300X 192GBCheapestCDNA 3 · HBM3$2.393sold as 4 GPUs; you need 3$9.56$6,979RunPod (opens in a new tab)4x MI300X
26 × RTX PRO 6000 96GBBlackwell · GDDR7$1.806sold as 8 GPUs; you need 6$14.40$10,512Nebius (opens in a new tab)8x RTX PRO 6000 96GB
34 × H200 141GBHopper · HBM3e$3.794$15.16$11,067RunPod (opens in a new tab)4x H200 NVL
48 × A100 80GBAmpere · HBM2e$2.098$16.72$12,206RunPod (opens in a new tab)8x A100 PCIe
52 × B300 288GBBlackwell Ultra · HBM3e$8.992$17.98$13,125RunPod (opens in a new tab)2x B300
68 × H100 80GBHopper · HBM3$2.898$23.12$16,878RunPod (opens in a new tab)8x H100 PCIe
76 × H100 NVL 94GBHopper · HBM3$3.196sold as 8 GPUs; you need 6$25.52$18,630RunPod (opens in a new tab)8x H100 NVL
84 × B200 180GBBlackwell · HBM3e$7.794$31.16$22,747Lambda (opens in a new tab)4x B200 SXM6
94 × GB200 186GBBlackwell · HBM3e$10.504$42.00$30,660CoreWeave (opens in a new tab)GB200 NVL72
More than 8: needs several nodes: L40S 48GB, RTX 6000 Ada 48GB, RTX A6000 48GB, A40 48GB, A100 40GB, RTX 5090 32GB, RTX 4090 24GB, RTX 3090 24GB, L4 24GB, A10 24GB, T4 16GB, V100 16GB
Cost per cloudCheapest offer that fits, per provider. Approximate, prices collected Nov 16, 2026.
ProviderInstancePer GPU / hrGPUs neededTotal / hrTotal / month (24/7)
RunPod (opens in a new tab)Cheapest4x MI300Xus$2.393 × MI300X 192GB sold as 4 GPUs; you need 3about $9.56about $6,979
Nebius (opens in a new tab)8x RTX PRO 6000 96GBeu/us$1.806 × RTX PRO 6000 96GB sold as 8 GPUs; you need 6about $14.40about $10,512
Vast.ai (opens in a new tab)8x RTX PRO 6000 S (median of 4 offers)worldwide marketplace$2.236 × RTX PRO 6000 96GB sold as 8 GPUs; you need 6about $17.87about $13,045
CoreWeave (opens in a new tab)RTX PRO 6000 Blackwell Server Edition (High Memory)us$2.506 × RTX PRO 6000 96GB sold as 8 GPUs; you need 6about $20.00about $14,600
Vultr (opens in a new tab)vbm-112c-2048gb-8-a100-gpu (bare metal)us$2.808 × A100 80GB about $22.40about $16,352
Lambda (opens in a new tab)8x HGX A100us$3.198 × A100 80GB about $25.52about $18,630
AWS (opens in a new tab)p4de.24xlargeus-east-1$3.438 × A100 80GB about $27.45about $20,036
Azure (opens in a new tab)ND96amsr_A100_v4eastus$4.108 × A100 80GB about $32.77about $23,922
Google Cloud (opens in a new tab)g4-standard-384us-central1$4.506 × RTX PRO 6000 96GB sold as 8 GPUs; you need 6about $36.00about $26,280

Prices collected Nov 16, 2026. Every cost is approximate: on-demand, US dollars, mostly US regions (marketplaces such as Vast.ai and Lium are worldwide).

Memory by precision and workload

With a 8K-token context and one request at a time. LoRA and full fine-tuning start from 16-bit weights and QLoRA from 4-bit, so the other cells are blank.
PrecisionInferenceLoRA fine-tuneQLoRA fine-tuneFull fine-tune
FP16 / BF16506 GB589 GB–4,030 GB
INT8267 GB–––
INT4135 GB–218 GB–

Cheapest GPU by precision

Inference with an 8K context and one request at a time.

FP16 / BF16

about 506 GB · 8K tokens

The weights as published. Nothing lost, but the largest footprint.

3 × MI300X 192GB

RunPod (opens in a new tab) · 4x MI300X

sold as 4 GPUs; you need 3
Per GPU / hr
$2.39
GPUs needed
3
Total / hr
about $9.56
per month, 24/7, approx.
$6,979

INT8

about 267 GB · 8K tokens

Weights at half the size. Most chat and RAG work shows no visible difference.

7 × A40 48GB

RunPod (opens in a new tab) · 8x A40

sold as 8 GPUs; you need 7
Per GPU / hr
$0.59
GPUs needed
7
Total / hr
about $4.72
per month, 24/7, approx.
$3,446

INT4

about 135 GB · 8K tokens

Weights at a quarter of the size. Small drops show up first in maths and code.

4 × A40 48GB

RunPod (opens in a new tab) · 4x A40

Per GPU / hr
$0.59
GPUs needed
4
Total / hr
about $2.36
per month, 24/7, approx.
$1,723

Inference vs fine-tuning

  • Inference (FP16)

    1×

    of the memory inference needs

    506 GB

    3 × MI300X 192GB · about $9.56

  • LoRA fine-tune

    1.2×

    of the memory inference needs

    589 GB

    4 × MI300X 192GB · about $9.56

  • QLoRA fine-tune

    0.4×

    of the memory inference needs

    218 GB

    6 × A40 48GB · about $4.72

  • Full fine-tune

    8×

    of the memory inference needs

    4,030 GB

    more than one 8-GPU node

Serving it at FP16 fits on 3 × MI300X 192GB (RunPod). LoRA needs 4 × MI300X 192GB (RunPod), QLoRA 6 × A40 48GB (RunPod) and a full fine-tune more than one 8-GPU node. Training keeps activations and optimiser state in memory as well as the weights, which is why it so often needs more GPUs than serving the same model.

Will it run on a Mac or a single RTX 4090?

RTX 4090 (24 GB)

Not on its own. Even at INT4 it needs about 135 GB, more than the card’s 24 GB. Two or more cards, or a bigger data-centre GPU, will do it.

Mac with Apple silicon

Yes, with enough memory. At INT4 it needs about 135 GB, so a Mac with 192 GB or more of unified memory (macOS lets the GPU use about three quarters of it). Expect slower generation than on a data-centre GPU.

Similar models

Open models of a similar size.

Calculate another model

MiniMax M2.5: common questions

How much VRAM does MiniMax M2.5 need?

About 506 GB at FP16, 267 GB at INT8 and 135 GB at INT4, with an 8K context and one request at a time. Longer contexts and more parallel requests add KV cache on top.

What is the cheapest GPU to run MiniMax M2.5?

3 × MI300X 192GB (about $2.39 per GPU / hour; sold as 4 GPUs; you need 3) on RunPod, at about $9.56 an hour for FP16 / BF16 inference with an 8K context. Prices collected Nov 16, 2026; they change often.

Can I fine-tune MiniMax M2.5 on a single GPU?

Not on one GPU. Even QLoRA needs about 218 GB, more than an 80 GB GPU holds. LoRA needs about 589 GB and a full fine-tune about 4,030 GB.

How are these numbers calculated?

Weights are parameters times bytes per parameter (2 at FP16, 1 at INT8, 0.5 at INT4). The KV cache comes from the model’s layers, KV heads, head size and context length. We add about 10% plus 1 GB for the runtime and fill each GPU to 90%. Prices are on-demand list prices in US regions, collected from each provider’s public pricing.

Can MiniMax M2.5 run on a consumer GPU like the RTX 4090?

Not on its own. Even at INT4 it needs about 135 GB, more than the card’s 24 GB. Two or more cards, or a bigger data-centre GPU, will do it.

Does MiniMax M2.5 fit on a single H100 80GB?

No. Even at INT4 it needs about 135 GB, so you need several GPUs.

How many GPUs do I need for MiniMax M2.5?

At FP16 with an 8K context: 8 × 80 GB, 12 × 48 GB or 24 × 24 GB GPUs. At INT4 you need about a quarter of the memory for the weights.

How much VRAM does MiniMax M2.5 need at INT8?

About 267 GB: 240 GB for the weights, 2.1 GB for an 8K KV cache and 25 GB of runtime overhead.

How much does it cost to run MiniMax M2.5 24/7 for a month?

About $6,979 on the cheapest setup we found (3 × MI300X 192GB (about $2.39 per GPU / hour; sold as 4 GPUs; you need 3), $9.56 an hour for 730 hours). The cheapest on AWS, Google Cloud or Azure is about $20,036 a month.

FP16 vs INT8 vs INT4: which should I use?

FP16 keeps full quality and needs the most memory (about 506 GB here). INT8 halves the weights with little loss for most chat work. INT4 (AWQ, GPTQ or GGUF Q4) cuts them to a quarter, about 135 GB in total, with a small drop in quality that shows most in maths and code.

Can MiniMax M2.5 run on a Mac?

Yes, with enough memory. At INT4 it needs about 135 GB, so a Mac with 192 GB or more of unified memory (macOS lets the GPU use about three quarters of it). Expect slower generation than on a data-centre GPU.

vLLM or llama.cpp?

vLLM is built for serving many requests at once on NVIDIA and AMD GPUs and reserves most of the GPU memory for KV cache. llama.cpp runs GGUF files on almost anything, including Macs and CPUs, and suits one user or a small team. Memory needs are similar for the same precision.

Prices collected Nov 16, 2026. Every cost is approximate: on-demand, US dollars, mostly US regions (marketplaces such as Vast.ai and Lium are worldwide).

How we calculate

  • Weights. Parameters times bytes per parameter: 2 at FP16, 1 at INT8, 0.5 at INT4, plus about 5% for the quantisation scales at INT8 and INT4.
  • KV cache. 2 × layers × KV heads × head size × context × requests × 2 bytes, from the model’s config.json. Where a setting is missing we estimate it and say so.
  • Overhead. About 10% plus 1 GB for the runtime, in line with what vLLM reserves.
  • Fine-tuning. LoRA keeps 16-bit weights and adds a small adapter, its optimiser and activations. QLoRA does the same over 4-bit weights. A full fine-tune needs about 16 bytes per parameter with mixed-precision Adam (weights, gradients and optimiser states), plus activations.
  • GPUs. We fill each GPU to 90% of its memory and stop at 8 per node. Costs are on-demand list prices in US regions; marketplace clouds show the median offer.