MiniMax · Over 80B · Mixture of experts
MiniMax M2.5
228.7 billion parameters in total, but only 10.98 billion are read for each token. That split is the whole point of the design: it costs the memory of a large model and the speed of a small one. It uses grouped-query attention with 8 KV heads for 48 query heads, which cuts the cache to 17% of what full attention would need.
About MiniMax M2.5
MiniMax M2.5 is a mixture-of-experts model with 228.7B parameters, 10.98B of them active for each token, built from 62 layers with a hidden size of 3072. Its repository was published on 2026-02-12. The license listed on Hugging Face is other. It uses grouped-query attention with 8 KV heads for 48 query heads, which cuts the cache to 17% of what full attention would need.
Memory needed, by quantization
At an 8k context window. Quality is our rough ranking of how much the compression costs you: anything at or above Q5 is hard to tell apart from the original in normal use.
| Quantization | File size | KV cache | Total VRAM | Quality | What it costs you |
|---|---|---|---|---|---|
| Q8_0 | 226.31 GB | 1.94 GB | 228.01 GB | 99% | Lossless in practice. Use it when the memory is there. |
| Q6_K | 174.65 GB | 1.94 GB | 176.5 GB | 98% | Very close to Q8 for two thirds of the size. |
| Q5_K_M | 151.49 GB | 1.94 GB | 153.39 GB | 96% | The quality-first choice when Q6 will not fit. |
| Q4_K_M | 128.59 GB | 1.94 GB | 130.56 GB | 93% | The default. Best size-to-quality ratio for local use. |
| IQ4_XS | 113.15 GB | 1.94 GB | 115.16 GB | 91% | Importance-matrix 4-bit. Q4_K_S quality, smaller file. |
| Q3_K_M | 104.1 GB | 1.94 GB | 106.13 GB | 86% | Degradation starts to show. A way to fit one size up. |
| IQ3_XXS | 81.47 GB | 1.94 GB | 83.56 GB | 79% | Aggressive. Only worth it on very large models. |
What a longer conversation costs
Same model at Q4_K_M, only the context window changes. The cache grows in a straight line with every token in the window.
| Context | KV cache | Total VRAM | Fits in 8 GB | Fits in 12 GB | Fits in 24 GB |
|---|---|---|---|---|---|
| 4k | 0.97 GB | 129.59 GB | no | no | no |
| 8k | 1.94 GB | 130.56 GB | no | no | no |
| 16k | 3.88 GB | 132.5 GB | no | no | no |
| 32k | 7.75 GB | 136.37 GB | no | no | no |
| 64k | 15.5 GB | 144.12 GB | no | no | no |
| 128k | 31 GB | 159.62 GB | no | no | no |
The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.
Which hardware runs MiniMax M2.5
1 of 118 consumer devices run it at a quantization worth using, at an 8k context window.
| Device | Memory | Verdict | Quantization | Used | Speed |
|---|---|---|---|---|---|
| Apple M3 Ultra 512GB Apple | 512 GB | Runs great | Q8_0 | 228.59 GB | 25.3 t/s |
| Apple M1 Ultra 64GB Apple | 64 GB | Just fits | n/a ! | 57.13 GB | 69.1 t/s |
| Apple M2 Ultra 64GB Apple | 64 GB | Just fits | n/a ! | 57.13 GB | 69.1 t/s |
| Apple M3 Ultra 96GB Apple | 96 GB | Just fits | n/a ! | 83.75 GB | 55.3 t/s |
| Apple M4 Max 64GB Apple | 64 GB | Just fits | n/a ! | 57.13 GB | 47.2 t/s |
| Apple M1 Ultra 128GB Apple | 128 GB | Just fits | n/a | 115.43 GB | 42.9 t/s |
| Apple M2 Ultra 128GB Apple | 128 GB | Just fits | n/a | 115.43 GB | 42.9 t/s |
| Apple M1 Max 64GB Apple | 64 GB | Just fits | n/a ! | 57.13 GB | 34.6 t/s |
| Apple M2 Max 64GB Apple | 64 GB | Just fits | n/a ! | 57.13 GB | 34.6 t/s |
| Apple M3 Max 64GB Apple | 64 GB | Just fits | n/a ! | 57.13 GB | 34.6 t/s |
| Apple M2 Ultra 192GB Apple | 192 GB | Just fits | n/a | 153.77 GB | 34.3 t/s |
| Apple M4 Max 128GB Apple | 128 GB | Just fits | n/a | 115.43 GB | 29.3 t/s |
| Apple M2 Max 96GB Apple | 96 GB | Just fits | n/a ! | 83.75 GB | 27 t/s |
| Apple M3 Max 96GB Apple | 96 GB | Just fits | n/a ! | 83.75 GB | 27 t/s |
| Apple M3 Ultra 256GB Apple | 256 GB | Just fits | n/a | 228.59 GB | 25.3 t/s |
| Apple M4 Pro 64GB Apple | 64 GB | Just fits | n/a ! | 57.13 GB | 23.6 t/s |
| Apple M3 Max 128GB Apple | 128 GB | Just fits | n/a | 115.43 GB | 21.4 t/s |
| Jetson AGX Orin 64GB NVIDIA | 64 GB | Just fits | n/a ! | 57.16 GB | 19.2 t/s |
| Ryzen AI Max+ 395 64GB AMD | 64 GB | Just fits | n/a ! | 57.16 GB | 18.5 t/s |
| NVIDIA DGX Spark 128GB NVIDIA | 128 GB | Just fits | n/a | 115.46 GB | 15.8 t/s |
| Ryzen AI Max+ 395 96GB AMD | 96 GB | Just fits | n/a ! | 83.78 GB | 14.4 t/s |
| Ryzen AI Max+ 395 128GB AMD | 128 GB | Just fits | n/a | 115.46 GB | 11.5 t/s |
| RTX PRO 6000 Blackwell NVIDIA | 96 GB | Partial offload 62/62 | n/a | 130.56 GB | 21.9 t/s |
| RTX A6000 NVIDIA | 48 GB | Partial offload 62/62 | n/a | 115.16 GB | 11.6 t/s |
| Radeon PRO W7900 AMD | 48 GB | Partial offload 62/62 | n/a | 115.16 GB | 11.5 t/s |
Show 35 more devices
| GeForce RTX 5090 NVIDIA | 32 GB | Partial offload 62/62 | n/a | 115.16 GB | 10.7 t/s |
| GeForce RTX 4090 NVIDIA | 24 GB | Partial offload 62/62 | n/a | 106.13 GB | 10.4 t/s |
| GeForce RTX 3090 Ti NVIDIA | 24 GB | Partial offload 62/62 | n/a | 106.13 GB | 10.4 t/s |
| GeForce RTX 3090 NVIDIA | 24 GB | Partial offload 62/62 | n/a | 106.13 GB | 10.3 t/s |
| GeForce RTX 5090 Laptop NVIDIA | 24 GB | Partial offload 62/62 | n/a | 106.13 GB | 10.3 t/s |
| Radeon RX 7900 XTX AMD | 24 GB | Partial offload 62/62 | n/a | 106.13 GB | 10.2 t/s |
| RTX A5000 NVIDIA | 24 GB | Partial offload 62/62 | n/a | 106.13 GB | 10.1 t/s |
| RTX 6000 Ada Generation NVIDIA | 48 GB | Partial offload 62/62 | n/a | 130.56 GB | 10 t/s |
| GeForce RTX 5080 NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.7 t/s |
| GeForce RTX 5070 Ti NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.6 t/s |
| Radeon RX 7900 XT AMD | 20 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.6 t/s |
| GeForce RTX 4080 SUPER NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.5 t/s |
| GeForce RTX 4080 NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.5 t/s |
| GeForce RTX 5080 Laptop NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.5 t/s |
| GeForce RTX 4070 Ti SUPER NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.4 t/s |
| GeForce RTX 3080 Ti NVIDIA | 12 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.4 t/s |
| GeForce RTX 3080 12GB NVIDIA | 12 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.4 t/s |
| GeForce RTX 4090 Laptop NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.3 t/s |
| GeForce RTX 5070 NVIDIA | 12 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.2 t/s |
| GeForce RTX 3080 10GB NVIDIA | 10 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.2 t/s |
| Radeon RX 9070 XT AMD | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.2 t/s |
| Radeon RX 9070 AMD | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.2 t/s |
| Radeon RX 7800 XT AMD | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.2 t/s |
| GeForce RTX 2080 Ti NVIDIA | 11 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.1 t/s |
| Radeon RX 7900 GRE AMD | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7.1 t/s |
| GeForce RTX 5060 Ti 16GB NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| GeForce RTX 4070 Ti NVIDIA | 12 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| GeForce RTX 4070 SUPER NVIDIA | 12 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| GeForce RTX 4070 NVIDIA | 12 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| RTX A4000 NVIDIA | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| Radeon RX 6900 XT AMD | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| Radeon RX 6800 AMD | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| Arc A770 16GB Intel | 16 GB | Partial offload 62/62 | n/a | 130.56 GB | 7 t/s |
| GeForce RTX 3070 Ti NVIDIA | 8 GB | Partial offload 62/62 | n/a | 130.56 GB | 6.9 t/s |
| GeForce GTX 1080 Ti NVIDIA | 11 GB | Partial offload 62/62 | n/a | 130.56 GB | 6.9 t/s |
How it compares
Other models in the same family or of a similar size, on a 12 GB card and a 24 GB card, each at the best quantization that fits.
| Model | Weights at 4 bits | GeForce RTX 3060 12GB | GeForce RTX 4090 |
|---|---|---|---|
| MiniMax M2.5 | 128.6 GB | does not fit well | 10.4 t/s at Q3_K_M |
| MiniMax M2.7 | 128.6 GB | does not fit well | 10.4 t/s at Q3_K_M |
| MiniMax M3 | 240.1 GB | does not fit well | does not fit well |
| Qwen3 235B A22B | 132.2 GB | does not fit well | does not fit well |
What to buy to run MiniMax M2.5
The cheapest hardware that runs it at a quantization worth using and a speed you would not resent, at an 8k context window.
Cheapest that works
Apple M1 Ultra 128GB
$3,400 whole machine
IQ4_XS · uses 115.43 GB of its 128 GB · about 42.9 tokens/s
Best value
NVIDIA DGX Spark 128GB
$3,999 whole machine
IQ4_XS · about 15.8 tokens/s · 273 GB/s
If it just has to run
GeForce GTX 1660 SUPER
$110 used
Q4_K_M · about 6.3 tokens/s. Cheaper, and slower to answer.
| Device | Price | Memory | Runs it at | Speed | |
|---|---|---|---|---|---|
| Apple M1 Ultra 128GB Apple · whole machine | $3,400 | 128 GB | IQ4_XS | 42.9 t/s | Check price |
| Apple M3 Ultra 256GB Apple · whole machine | $5,599 | 256 GB | Q8_0 | 25.3 t/s | Check price |
| Apple M3 Ultra 512GB Apple · whole machine | $9,499 | 512 GB | Q8_0 | 25.3 t/s | Check price |
Indicative prices reviewed 2026-09-01; used prices are marketplace typical. Options that cost more than a cheaper one with no more memory and no more speed are hidden. Price links are Amazon affiliate links. Change the standard, the context window or the budget in the buying tool.
Where it sits in the catalog
Source: MiniMaxAI/MiniMax-M2.5 on Hugging Face. Downloaded 487k times in the last month. Published 2026-02-12. Architecture figures are read from the repository's own configuration file, so they move when the model does.
Questions people ask about MiniMax M2.5
How much VRAM does MiniMax M2.5 need?
At Q4_K_M with an 8k context window, MiniMax M2.5 needs about 130.56 GB: 128.25 GB of weights, 1.94 GB of KV cache and 0.37 GB of overhead. At the highest-quality file it needs about 228.01 GB.
What is the cheapest GPU that runs MiniMax M2.5?
The smallest consumer device in our catalog that fits it is the Apple M3 Ultra 512GB with 512 GB, running it at Q8_0 at roughly 25.3 tokens per second.
How much does a longer context window add for MiniMax M2.5?
On the memory side, a 32k window adds 7.8 GB of KV cache on top of the weights, and the cache grows in a straight line with every token. The longest window the model supports natively is 192k tokens.