- Services
Every open model in our directory, ranked by how well it runs on NIMO Mini PC (Ryzen AI Max+ 395 128GB) (96GB) at 4-bit. Grades weigh whether the model fits in memory and how fast it should run.
NIMO Mini PC (Ryzen AI Max+ 395 128GB) has 96 GB of VRAM, enough to run 264 of the 298 open models we track at 4-bit. The largest that still fits well is Kimi K2 Instruct (1000B), and the top-rated is Holo4-35B-A3B at about 88 tok/s.
264 of 298 models run comfortably on NIMO Mini PC (Ryzen AI Max+ 395 128GB).
See full NIMO Mini PC (Ryzen AI Max+ 395 128GB) specs and pricingThe top models this device can run at 4-bit, ranked by fit and speed.
| Model | Grade | Speed | VRAM |
|---|---|---|---|
| Holo4-35B-A3BHcompany | AA | 88.1 tok/s | 2.3 GB |
| Qwen3-30B-A3BAlibaba | AA | 38.3 tok/s | 5.4 GB |
| Mellum2.1JetBrains | AA | 101.4 tok/s | 2.0 GB |
| LFM2.5-8B-A1BLiquid AI | AA | 70.9 tok/s | 2.9 GB |
| Apertus 8BEPFL, ETH Zurich, CSCS | AA | 38.1 tok/s | 5.4 GB |
| Llama 3 8B InstructMeta | AA | 36.4 tok/s | 5.7 GB |
| PersonaPlex 7BNVIDIA | AA | 43.0 tok/s | 4.8 GB |
| Llama 2 7B ChatMeta | AA | 43.0 tok/s | 4.8 GB |
| LensVLM-9BApple | AA | 34.3 tok/s | 6.0 GB |
| Carnice-9b for Hermes agentkai-os | AA | 34.3 tok/s | 6.0 GB |
| MiniCPM5-2Bopenbmb | AA | 44.8 tok/s | 4.6 GB |
| LFM2.5-2.6BLiquid AI | AA | 42.2 tok/s | 4.9 GB |
Holo4-35B-A3BHcompany | 35B(3B active) | AA | 88.1 tok/s | 2.3 GB | |
Qwen3-30B-A3BAlibaba | 30B(3B active) | AA | 38.3 tok/s | 5.4 GB | |
Mellum2.1JetBrains | 12B(2.5B active) | AA | 101.4 tok/s | 2.0 GB | |
LFM2.5-8B-A1BLiquid AI | 8.3B(1.5B active) | AA | 70.9 tok/s | 2.9 GB | |
Apertus 8BEPFL, ETH Zurich, CSCS | 8B | AA | 38.1 tok/s | 5.4 GB | |
| 8B | AA | 36.4 tok/s | 5.7 GB | ||
PersonaPlex 7BNVIDIA | 7B | AA | 43.0 tok/s | 4.8 GB | |
Llama 2 7B ChatMeta | 7B | AA | 43.0 tok/s | 4.8 GB | |
Ad Tools for AI BuildersAll the hardware and tools you need to run private AI at your office or at home.Shop Now | |||||
LensVLM-9BApple | 9B | AA | 34.3 tok/s | 6.0 GB | |
| 9B | AA | 34.3 tok/s | 6.0 GB | ||
MiniCPM5-2Bopenbmb | 2.52B | AA | 44.8 tok/s | 4.6 GB | |
LFM2.5-2.6BLiquid AI | 2.69B | AA | 42.2 tok/s | 4.9 GB | |
VibeThinker-3BWeiboAI | 3B | AA | 54.0 tok/s | 3.8 GB | |
Gemma 4 E2B ITGoogle | 2B | AA | 55.6 tok/s | 3.7 GB | |
| 81.29B(3B active) | AA | 24.2 tok/s | 8.5 GB | ||
North Mini CodeCohere | 30B(3B active) | AA | 24.6 tok/s | 8.4 GB | |
Ad Vast.aiAffordable on-demand GPU rentals for training and inference. Pick from thousands of hosts.Rent a GPU | |||||
Nemotron 3 Nano OmniNVIDIA | 30B(3B active) | AA | 24.2 tok/s | 8.5 GB | |
Qwen3.6 35B-A3BAlibaba | 35B(3B active) | AA | 24.2 tok/s | 8.5 GB | |
Qwen3.5-35B-A3BAlibaba | 35B(3B active) | AA | 24.2 tok/s | 8.5 GB | |
Youtu-Parsing-Omnitencent | 5.33B | AA | 32.3 tok/s | 6.4 GB | |
Mistral 7B InstructMistral AI | 7B | AA | 32.2 tok/s | 6.4 GB | |
Llama 2 13B ChatMeta | 13B | AA | 24.3 tok/s | 8.5 GB | |
Gemma 4 E4B ITGoogle | 4B | BB | 29.8 tok/s | 6.9 GB | |
Gemma 3 4B ITGoogle | 4B | BB | 29.8 tok/s | 6.9 GB | |
| AdTools for AI BuildersAll the hardware and tools you need to run private AI at your office or at home.Shop Now | |||||
BolmoAi2 | 0B | BB | 412.2 tok/s | 0.5 GB | |
Model Compatibility
Use the calculator to weigh this device against any other, model by model, with speed and fit scores.
Straight answers to the questions we hear most often.
Can't find what you're looking for? Book a Discovery Call
Holo4-35B-A3B is the highest-rated model that runs on NIMO Mini PC (Ryzen AI Max+ 395 128GB) at 4-bit, at about 88 tok/s. In total, NIMO Mini PC (Ryzen AI Max+ 395 128GB) can run 264 of the 298 open models we track.
We size each model at 4-bit (Q4_K_M), add the KV cache for its context length and a fixed runtime overhead, then compare that to this device VRAM. Each model is graded on how comfortably it fits and how fast it should run on this card.
It is an estimate of decode speed in tokens per second at 4-bit, based on this device memory bandwidth and each model size. Real speed depends on your runtime, batch size, and context length, so use it to compare models, not as a hard promise.
Lower-bit quantization shrinks a model so more of them fit. These grades already assume 4-bit, which is the common local default. Going below 4-bit can squeeze a larger model in at some quality cost; going above needs more VRAM and may not fit.
Mixture-of-experts models only activate a fraction of their parameters at a time, so their runtime memory is closer to the active size than the headline size. We score fit on the active parameters, which is why some very large models still fit.