AI mini PCs
Which AI models run on the DGX Spark (GB10, 128 GB)?
The DGX Spark (GB10, 128 GB) has 128 GB of memory, about 122 GB of it usable for a model, and 273 GB/s of memory bandwidth. 103 of the 110 models in our catalog run on it at 4-bit or better.
Sold as: NVIDIA DGX Spark, ASUS Ascent GX10, Dell Pro Max with GB10, Lenovo ThinkStation PGX, MSI EdgeXpert, HP ZGX Nano G1n, Acer Veriton GN100, Gigabyte AI TOP ATOM
Data as of 2 November 2026 · Vendor specs
128 GBmemory
~122 GB usable for a model
273 GB/sbandwidth, sets the speed
103models run at 4-bit or better
Also covers other GB10 systems with 128 GB (ASUS Ascent GX10, Dell Pro Max GB10, Lenovo ThinkStation PGX, MSI EdgeXpert). Usable figure: addressable unified memory.
Our pick for this machine Qwen3.8-Flash-Next The largest model on our shortlist of current models that runs at 4-bit or better and 10 tok/s or more.Models that run on it
| Model | Best build | Memory | Speed |
|---|---|---|---|
| Qwen3.8-Flash-Next125B MoE | Q4_K_XL | 115 GB | ~28 tok/s our est. |
| Nemotron 3 Super 120B-A12B120B MoE | Q4_K_M | 90 GB | ~23 tok/s reported |
| Nemotron 3.5 Super VL 120B-A12B120B MoE | Q6_K | 102 GB | ~15 tok/s our est. |
| gpt-oss-120b117B MoE | MXFP4 | 68 GB | ~61 tok/s reported |
| Llama 3.3 70B70.6B | Q4_K_M | 44 GB | ~6.0 tok/s slow measured |
| Llama 3.3 Nemotron Super 49B49B | Q4_K_M | 31 GB | ~5.1 tok/s slow our est. |
| Qwen3.6-35B-A3B35B MoE | Q8_0 | 41 GB | ~45 tok/s our est. |
| Gemma 4 31B30.7B | Q4_K_M | 21 GB | ~7.5 tok/s slow our est. |
| Qwen3-Coder-30B-A3B30.5B MoE | Q8_0 | 34 GB | ~43 tok/s our est. |
| Nemotron 3 Nano 30B-A3B30B MoE | FP16 | 62 GB | ~25 tok/s our est. |
| Qwen3.8-27B27.3B | Q4_K_M | 17.9 GB | ~11 tok/s reported |
| Gemma 4 26B-A4B25.2B MoE | Q8_0 | 30 GB | ~52 tok/s reported |
| gpt-oss-20b21B MoE | MXFP4 | 15.1 GB | ~63 tok/s our est. |
| Nemotron Nano 9B v29B | Q8_0 | 10.8 GB | ~16 tok/s our est. |
| Qwen3.5 9B Instruct9B | Q8_0 | 12.0 GB | ~14 tok/s our est. |
| Qwen2.5 7B Instruct7B | Q4_K_M | 6.6 GB | ~27 tok/s our est. |
| Gemma 3 4B Instruct4B | Q8_0 | 5.0 GB | ~39 tok/s our est. |
| Ornith 1.5 122B125.09B | Q4_K_M | 79 GB | no estimate |
| Ling-3.0-flash124B | IQ4_XS | 71 GB | no estimate |
| Devstral 2 123B123B | Q4_K_M | 78 GB | no estimate |
| Qwen3.5 122B-A10B Instruct122B MoE | Q4_K_M | 75 GB | ~25 tok/s our est. |
| Laguna S 2.1118B | Q4_K_M | 99 GB | no estimate |
| Llama 4 Scout109B MoE | Q4_K_M | 70 GB | ~14 tok/s our est. |
| Qwen3-Next 80B-A3B80B MoE | Q8_0 | 88 GB | ~47 tok/s our est. |
| Kolibri 178.1B | IQ4_XS | 46 GB | no estimate |
| Llama 3.1 70B Instruct70B | Q4_K_M | 44 GB | ~3.6 tok/s slow our est. |
| DeepSeek-R1 Distill Llama 70B70B | Q4_K_M | 44 GB | ~3.6 tok/s slow our est. |
| Mixtral 8x7B Instruct46.7B MoE | Q4_K_M | 32 GB | ~18 tok/s our est. |
| Flint 40B-A2.5B39.8B MoE | NVFP4 | 24 GB | ~104 tok/s our est. |
| Sage-A1.535.95B | Q4_K_M | 23 GB | no estimate |
| Qwen3.5 35B-A3B Instruct35B MoE | Q8_0 | 41 GB | ~45 tok/s our est. |
| Ornith 1.0 35B35B | Q4_K_M | 23 GB | no estimate |
| Nanbeige4.3 33B33.1B | Q4_K_M | 22 GB | no estimate |
| Poolside Laguna XS.233B | Q4_K_M | 25 GB | no estimate |
| Laguna XS 2.133B | Q4_K_M | 22 GB | no estimate |
| Qwen3 32B32B | Q4_K_M | 22 GB | ~7.4 tok/s slow our est. |
| DeepSeek-R1 Distill Qwen 32B32B | Q4_K_M | 21 GB | ~7.5 tok/s slow our est. |
| Olmo 3.1 32B Instruct32B | Q4_K_M | 21 GB | no estimate |
| Qwen3 30B30B | Q4_K_M | 24 GB | ~6.8 tok/s slow our est. |
| NVIDIA Nemotron Cascade 2 30B-A3B30B MoE | Q6_K | 26 GB | ~63 tok/s our est. |
| Cohere North Mini Code30B | Q4_K_M | 20 GB | no estimate |
| Nemotron 3.5 Lightning 30B-A3B30B MoE | Q4_K_M | 27 GB | ~59 tok/s our est. |
| GLM-4.7-Flash 30B-A3B30B MoE | Q4_K_M | 20 GB | ~79 tok/s our est. |
| Granite Switch 4.1 30B30B | Q4_K_M | 23 GB | no estimate |
| Muse Glimmer 30B29.8B | Q4_K_M | 18.3 GB | no estimate |
| Granite 4.2 30B29.3B | Q4_K_M | 17.9 GB | no estimate |
| AREX-2 27B27.4B | Q4_K_M | 18.6 GB | no estimate |
| Gemma 2 27B Instruct27B | Q4_K_M | 23 GB | ~7.2 tok/s slow our est. |
| Gemma 3 27B Instruct27B | Q4_K_M | 23 GB | ~7.2 tok/s slow our est. |
| Qwen3.5 27B Instruct27B | Q4_K_M | 17.4 GB | ~9.4 tok/s slow our est. |
| Qwen3.6 27B27B | Q4_K_M | 19.4 GB | ~8.3 tok/s slow our est. |
| Jebadiah 27B27B | Q4_K_M | 18.2 GB | no estimate |
| Mistral Small 3.124B | Q4_K_M | 16.4 GB | ~10 tok/s our est. |
| LFM2 24B-A2B Instruct24B MoE | Q4_K_M | 15.3 GB | ~129 tok/s our est. |
| Devstral Small 2 24B24B | Q4_K_M | 16.6 GB | no estimate |
| Mistral Small 22B22B | Q4_K_M | 18.4 GB | ~8.8 tok/s slow our est. |
| Qwen2.5 14B Instruct14B | Q4_K_M | 12.3 GB | ~14 tok/s our est. |
| Phi-3 Medium 14B14B | Q4_K_M | 12.3 GB | ~14 tok/s our est. |
| Phi-4 14B14B | Q8_0 | 15.9 GB | ~10 tok/s our est. |
| Qwen2.5 Coder 14B14B | Q4_K_M | 12.3 GB | ~14 tok/s our est. |
| Qwen3 14B14B | Q4_K_M | 12.3 GB | ~14 tok/s our est. |
| DeepSeek-R1 Distill Qwen 14B14B | Q8_0 | 16.0 GB | ~10 tok/s our est. |
| Ministral 3 14B13.5B | Q4_K_M | 10.3 GB | no estimate |
| Mellum2.1 12B-A2.5B Thinking12.1B MoE | Q4_K_M | 9.3 GB | ~90 tok/s our est. |
| Mistral Nemo 12B12B | Q4_K_M | 10.7 GB | no estimate |
| Gemma 3 12B Instruct12B | Q4_K_M | 10.7 GB | ~16 tok/s our est. |
| Gemma 4 12B12B | Q8_0 | 14.1 GB | ~12 tok/s our est. |
| GLM-4-9B-Chat9.4B | Q4_K_M | 7.5 GB | no estimate |
| Gemma 2 9B Instruct9B | Q4_K_M | 8.2 GB | ~21 tok/s our est. |
| Ornith 1.0 9B9B | Q4_K_M | 6.7 GB | no estimate |
| MiMo-V2.6-Distill-Qwen-9B9B | Q4_K_M | 6.9 GB | no estimate |
| Jebadiah 9B9B | Q4_K_M | 6.9 GB | no estimate |
| Granite 4.2 8B8.8B | Q4_K_M | 6.1 GB | no estimate |
| Granite Switch 4.1 8B8.8B | Q4_K_M | 7.6 GB | no estimate |
| MiniCPM-V 4.5 8B8.7B | Q4_K_M | 6.8 GB | no estimate |
| Ministral 3 8B8.4B | Q4_K_M | 7.1 GB | no estimate |
| LFM2.5 8B-A1B8.3B MoE | Q4_K_M | 6.6 GB | ~227 tok/s our est. |
| Llama 3.1 8B Instruct8B | Q8_0 | 9.2 GB | ~19 tok/s our est. |
| Qwen3 8B8B | Q8_0 | 9.3 GB | ~19 tok/s our est. |
| Granite 4.1 8B Instruct8B | Q4_K_M | 6.6 GB | no estimate |
| Mistral 7B Instruct7B | Q8_0 | 8.4 GB | ~21 tok/s our est. |
| Qwen2.5 Coder 7B7B | Q4_K_M | 6.6 GB | ~27 tok/s our est. |
| DeepSeek-R1 Distill Qwen 7B7B | Q8_0 | 8.7 GB | ~20 tok/s our est. |
| Gemma 4 E4B4.5B | Q4_K_M | 5.1 GB | no estimate |
| Qwen3.5 4B Instruct4B | Q8_0 | 5.4 GB | ~35 tok/s our est. |
| NVIDIA Nemotron 3.5 Bolt E4B4B | Q4_K_M | 9.1 GB | no estimate |
| Phi-3 Mini 3.8B3.8B | Q4_K_M | 4.3 GB | ~47 tok/s our est. |
| Phi-4 Mini 3.8B3.8B | Q8_0 | 4.9 GB | ~40 tok/s our est. |
| Granite 4.2 3B3.7B | Q4_K_M | 3.2 GB | no estimate |
| Ministral 3 3B3.4B | Q4_K_M | 4.1 GB | no estimate |
| Qwen2.5 3B Instruct3B | Q4_K_M | 3.6 GB | ~60 tok/s our est. |
| Llama 3.2 3B Instruct3B | Q4_K_M | 3.6 GB | ~60 tok/s our est. |
| Granite 4.1 3B Instruct3B | Q4_K_M | 3.0 GB | no estimate |
| Gemma 4 E2B2.3B | Q4_K_M | 3.4 GB | no estimate |
| Gemma 2 2B Instruct2B | Q4_K_M | 2.8 GB | ~83 tok/s our est. |
| Qwen3.5 2B Instruct2B | Q4_K_M | 2.8 GB | ~83 tok/s our est. |
| LFM3 1.7B1.73B | Q4_K_M | 2.1 GB | no estimate |
| Qwen2.5 1.5B Instruct1.5B | Q4_K_M | 2.5 GB | ~100 tok/s our est. |
| Llama 3.2 1B Instruct1B | Q4_K_M | 2.0 GB | ~150 tok/s our est. |
| Gemma 3 1B Instruct1B | Q8_0 | 2.0 GB | ~150 tok/s our est. |
| Qwen3.5 0.8B Instruct0.8B | Q4_K_M | 1.8 GB | ~188 tok/s our est. |
| Qwen2.5 0.5B Instruct0.5B | Q4_K_M | 1.8 GB | ~188 tok/s our est. |
| SmolLM2 360M0.36B | Q4_K_M | 1.5 GB | no estimate |
ModelFit does not cover this machine, so speeds come from our bandwidth formula; measured figures are on each model page where we have them. Method
Fits only at 3-bit or lower
Too big for this machine
Qwen3-235B-A22B (134 GB), GLM-5.3-Flash (200 GB), Llama 3.1 405B Instruct (250 GB), Llama 4 Maverick (252 GB), Ling-3.1-flash-1019 (368 GB), DeepSeek-R1 671B (390 GB).
Questions
What is the best AI model for the DGX Spark (GB10, 128 GB)?
Qwen3.8-Flash-Next at Q4_K_XL: the largest model on our shortlist of current models that runs at 4-bit or better at a usable speed (about 28 tok/s, our est.).
How much memory can a model use on the DGX Spark (GB10, 128 GB)?
About 122 GB of the 128 GB. Also covers other GB10 systems with 128 GB (ASUS Ascent GX10, Dell Pro Max GB10, Lenovo ThinkStation PGX, MSI EdgeXpert). Usable figure: addressable unified memory.
How fast is the DGX Spark (GB10, 128 GB) for local LLMs?
Decode speed is limited by memory bandwidth: 273 GB/s. A model that reads 10 GB per token runs at under 27 tokens per second.