With 128 GB unified memory (about 108.8 GB usable by the GPU), these are the best open models that fit. The top pick is gpt-oss 120B (MoE, MXFP4) at MXFP4 (~67.3 GB, ~55 tok/s): ollama run gpt-oss:120b.
Best for coding: gpt-oss 120B (MoE, MXFP4) (gpt-oss:120b), Qwen3-Coder 30B-A3B (MoE) (qwen3-coder:30b-a3b-q8_0), Qwen3 30B-A3B (MoE) (qwen3:30b-a3b-q8_0).
Runtimes available for this hardware: Ollama, LM Studio, llama.cpp, vLLM.
Open the interactive checker for DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte)| # | Model | Quant | Memory needed | Fit | Speed | Command |
|---|---|---|---|---|---|---|
| 1 | gpt-oss 120B (MoE, MXFP4) | MXFP4 | 67.3 GB | fits in fast memory | ~55 tok/s | ollama run gpt-oss:120b |
| 2 | Llama 4 Scout 109B (MoE) | Q4_K_M | 69.3 GB | fits in fast memory | ~15 tok/s | ollama run llama4:16x17b |
| 3 | Qwen3-Next 80B-A3B (MoE) | Q8_0 | 87.3 GB | fits in fast memory | ~49 tok/s | ollama run qwen3-next:80b-a3b-instruct-q8_0 |
| 4 | Qwen3-Coder 30B-A3B (MoE) | Q8_0 | 33.4 GB | fits in fast memory | ~45 tok/s | ollama run qwen3-coder:30b-a3b-q8_0 |
| 5 | Qwen3 30B-A3B (MoE) | Q8_0 | 34.4 GB | fits in fast memory | ~44 tok/s | ollama run qwen3:30b-a3b-q8_0 |
| 6 | gpt-oss 20B (MoE, MXFP4) | MXFP4 | 15.1 GB | fits in fast memory | ~65 tok/s | ollama run gpt-oss:20b |
| 7 | Mixtral 8x7B (MoE) | Q8_0 | 51.8 GB | fits in fast memory | ~11 tok/s | ollama run mixtral:8x7b-instruct-v0.1-q8_0 |
| 8 | Qwen3 14B | Q8_0 | 17.3 GB | fits in fast memory | ~13 tok/s | ollama run qwen3:14b-q8_0 |
| 9 | DeepSeek R1 Distill Qwen 14B | Q8_0 | 17.3 GB | fits in fast memory | ~13 tok/s | ollama run deepseek-r1:14b-qwen-distill-q8_0 |
| 10 | Phi-4 14B | Q8_0 | 17.3 GB | fits in fast memory | ~13 tok/s | ollama run phi4:14b-q8_0 |
| 11 | Mistral Small 24B | Q8_0 | 26.8 GB | fits in fast memory | ~8 tok/s | ollama run mistral-small:24b-instruct-2501-q8_0 |
| 12 | Mistral Nemo 12B | Q8_0 | 14.2 GB | fits in fast memory | ~16 tok/s | ollama run mistral-nemo:12b-instruct-2407-q8_0 |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-14.
A project by OnDevice Labs · llms.txt