What people run, by hardware
Pick the kind of machine you have. You'll see the model sizes people report running on it, most reported first, with the speed they typically get. It tells you what works on hardware like yours. It doesn't tell you which model gives the best answers.
To check a specific model on your exact card, use the calculator. For coding, see the coding models by VRAM.
Strix Halo and DGX Spark
128 GB of unified memory at about 260 GB/s: big models fit, and speed is set by that bandwidth. Memory speed 256 to 273 GB/s across the 4 devices with reports, so a typical figure here spans all of them. 260 single-device reports.
| # | Model | Reports | Gen t/s | PP t/s | With speculation |
|---|---|---|---|---|---|
| 1 | Qwen3.8 Flash-Next125B · 6B active Alibaba · mostly UD-Q4_K_XL Gen 35 t/s (17 runs) · PP 1200 t/s (15 runs) · with speculation 42 t/s (25) | 74 | 35 17 runs, 3 devices | 1200 15 runs | 42 25 runs |
| 2 | Qwen3.827B Alibaba · mostly NVFP4 Gen 9.5 t/s (10 runs) · PP 440 t/s (9 runs) · with speculation 40 t/s (14) | 37 | 9.5 10 runs, 2 devices | 440 9 runs | 40 14 runs |
| 3 | Qwen3.635B · 3B active Alibaba · mostly Q4_K_M Gen 52 t/s (16 runs) · PP 1017 t/s (14 runs) · with speculation 109 t/s (2) | 22 | 52 16 runs, 2 devices | 1017 14 runs | 109 2 runs |
| 4 | DeepSeek V4 Flash284B · 13B active DeepSeek · mostly IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8 Gen 26 t/s (2 runs) · PP 238 t/s (1 run) · with speculation 29 t/s (5) | 17 | 26 2 runs, 2 devices | 238 1 run | 29 5 runs |
| 5 | Qwen3-Coder30B · 3B active mostly UD-Q4_K_XL Gen 68 t/s (8 runs) · PP 402 t/s (7 runs) | 11 | 68 8 runs, 2 devices | 402 7 runs | none |
| 6 | Qwen3.627B Alibaba · mostly NVFP4 Gen 12 t/s (2 runs) · PP 322 t/s (1 run) · with speculation 21 t/s (3) | 10 | 12 2 runs | 322 1 run | 21 3 runs |
| 7 | Gemma 426B · 4B active Google DeepMind · mostly NVFP4 Gen 53 t/s (5 runs) · PP 3324 t/s (4 runs) · with speculation 102 t/s (1) | 8 | 53 5 runs, 2 devices | 3324 4 runs | 102 1 run |
| 8 | gpt-oss117B · 5.1B active OpenAI · mostly MXFP4 Gen 53 t/s (7 runs) · PP 645 t/s (7 runs) | 8 | 53 7 runs, 3 devices | 645 7 runs | none |
| 9 | Qwen3.535B · 3B active Alibaba · mostly UD-Q4_K_XL Gen 50 t/s (3 runs) · PP 1017 t/s (1 run) | 5 | 50 3 runs | 1017 1 run | none |
| 10 | Gemma 431B Google DeepMind · mostly UD-Q8_K_XL Gen 6.4 t/s (5 runs) · PP 530 t/s (4 runs) | 5 | 6.4 5 runs, 2 devices | 530 4 runs | none |
| 11 | Llama 3.23B Meta · mostly BF16 Gen 77 t/s (4 runs) · PP 1725 t/s (4 runs) | 4 | 77 4 runs | 1725 4 runs | none |
| 12 | Qwen330B · 3B active Alibaba · mostly Q4_K_M Gen 74 t/s (3 runs) · PP 920 t/s (3 runs) | 4 | 74 3 runs | 920 3 runs | none |
| 13 | gpt-oss21B · 3.6B active OpenAI · mostly MXFP4 Gen 72 t/s (4 runs) · PP 1089 t/s (4 runs) | 4 | 72 4 runs | 1089 4 runs | none |
| 14 | DeepSeek V4.1 Flash552B · 16B active mostly FP4 Gen 25 t/s (1 run) · PP 1228 t/s (1 run) | 4 | 25 1 run | 1228 1 run | none |
| 15 | Ornith1.535B · 3B active mostly NVFP4 Gen 71 t/s (2 runs) · PP 1505 t/s (1 run) · with speculation 79 t/s (1) | 3 | 71 2 runs, 2 devices | 1505 1 run | 79 1 run |
| 16 | Ling-3.0 Flash124B · 5.1B active Ant Group · mostly Q5_K_M Gen 39 t/s (3 runs) | 3 | 39 3 runs | no PP data | none |
| 17 | Qwen3.5122B · 10B active Alibaba · mostly UD-Q4_K_XL Gen 21 t/s (2 runs) · PP 287 t/s (1 run) | 3 | 21 2 runs | 287 1 run | none |
| 18 | Qwen3.54B Alibaba · mostly Q8_0 No generation data | 3 | no generation data | no PP data | none |
| 19 | Qwen3.59B Alibaba · mostly UD-Q4_K_XL Gen 31 t/s (1 run) · PP 819 t/s (1 run) | 2 | 31 1 run | 819 1 run | none |
| 20 | Nemotron 3 Super120B · 12B active mostly Q4_K Gen 17 t/s (2 runs) · PP 769 t/s (1 run) | 2 | 17 2 runs | 769 1 run | none |
| 21 | Qwen3.527B Alibaba · mostly UD-Q4_K_XL Gen 12 t/s (1 run) · PP 310 t/s (1 run) | 2 | 12 1 run | 310 1 run | none |
| 22 | Gemma 327B Google · mostly BF16 Gen 8.4 t/s (2 runs) · PP 243 t/s (2 runs) | 2 | 8.4 2 runs | 243 2 runs | none |
| 23 | Qwen332B Alibaba · mostly Q8_0 Gen 6.4 t/s (2 runs) · PP 68 t/s (2 runs) | 2 | 6.4 2 runs | 68 2 runs | none |
| 24 | GLM-5.3 Flash320B · 18B active Zhipu AI · mostly EXL3 No generation data | 2 | no generation data | no PP data | none |
| 25 | LFM2.52.6B Liquid AI · mostly Q4_K_M Gen 113 t/s (1 run) | 1 | 113 1 run | no PP data | none |
| 26 | Qwen3.50.8B Alibaba · mostly BF16 Gen 87 t/s (1 run) | 1 | 87 1 run | no PP data | none |
| 27 | Nex-N2.5-mini35B · 3B active mostly ROCmFP4 Gen 77 t/s (1 run) | 1 | 77 1 run | no PP data | none |
| 28 | Ornith 1.035B · 3B active mostly Q4_K_M Gen 74 t/s (1 run) · PP 1103 t/s (1 run) | 1 | 74 1 run | 1103 1 run | none |
| 29 | Ornith1.5 Ornith-1.5-35B-A3B35B · 3B active mostly Q6_K Gen 50 t/s (1 run) · PP 563 t/s (1 run) | 1 | 50 1 run | 563 1 run | none |
| 30 | Qwen3-Coder-Next80B · 3B active mostly UD-Q6_K_XL Gen 37 t/s (1 run) · PP 546 t/s (1 run) | 1 | 37 1 run | 546 1 run | none |
| 31 | Qwen3.8 Swift-1.5-125B-A6B125B · 6B active Alibaba · mostly IQ4_XS Gen 36 t/s (1 run) · PP 1634 t/s (1 run) | 1 | 36 1 run | 1634 1 run | none |
| 32 | Nemotron 3.5 Super VL120B · 12B active mostly Q4_K_M Gen 31 t/s (1 run) | 1 | 31 1 run | no PP data | none |
| 33 | MiniMax M2.7230B · 10B active mostly UD-IQ4_XS Gen 30 t/s (1 run) · PP 309 t/s (1 run) | 1 | 30 1 run | 309 1 run | none |
| 34 | Mimo 2.6 Flash-RL309B · 15B active Xiaomi · mostly IQ2_M Gen 26 t/s (1 run) · PP 310 t/s (1 run) | 1 | 26 1 run | 310 1 run | none |
| 35 | DeepSeek V4 Flash162B · 13B active DeepSeek · mostly NVFP4 Gen 14 t/s (1 run) · PP 539 t/s (1 run) | 1 | 14 1 run | 539 1 run | none |
| 36 | Llama 3.370B Meta · mostly Q4_K_XL Gen 5.1 t/s (1 run) · PP 78 t/s (1 run) | 1 | 5.1 1 run | 78 1 run | none |
| 37 | DeepSeek V4.1 Flash763B mostly IQ1_M No generation data | 1 | no generation data | no PP data | none |
| 38 | GLM-5.3 Flash / Maya-S-v2321B · 18B active Zhipu AI · mostly Maya-S-v2-IQ2_XXS No generation data | 1 | no generation data | no PP data | none |
| 39 | Gemma 4 E4B8B Google DeepMind · mostly Q4_K_M No generation data | 1 | no generation data | no PP data | none |
| 40 | Gemma 4 DiffusionGemma26B · 4B active Google DeepMind · mostly NVFP4 No generation data | 1 | no generation data | no PP data | none |
| 41 | Llama27B mostly Q4_0 No generation data | 1 | no generation data | no PP data | none |
| 42 | Mimo 2.6 Pro1020B · 42B active Xiaomi · mostly FP8 No generation data | 1 | no generation data | no PP data | none |
| 43 | Nemotron 3.5 Super VL120B · 12B active mostly NVFP4 No generation data · with speculation 28 t/s (1) | 1 | no generation data | no PP data | 28 1 run |
| 44 | Nemotron 3.5 Lightning30B · 3B active NVIDIA · mostly Q4_K_M No generation data | 1 | no generation data | no PP data | none |
| 45 | Qwen3.52B Alibaba · mostly Q8_0 No generation data | 1 | no generation data | no PP data | none |
| 46 | Qwen3.8 Flash-Next Uncensored (vmlinux Gufo Q4Mix)125B · 6B active Alibaba · mostly Q4 Mix (Q4_K/Q5_1/IQ4_NL/Q8_0) No generation data | 1 | no generation data | no PP data | none |
| 47 | Qwen3.8 Flash-Next Uncensored125B · 6B active Alibaba · mostly IQ4 No generation data | 1 | no generation data | no PP data | none |
Not in the table: 113 reports don't say which model size ran. 53 ran on more than one device, which says nothing about what one of them does.
How this list works
Each row is one model size, because a 27B and a 397B from the same family are different choices. Rows are ordered by how many people report running that size on this hardware.
Generation speed is the median of runs on one device, serving one request, without speculative decoding. PP is the median prompt-processing speed from plain runs that reported it, so its run count can be smaller. Speculative decoding (MTP, a draft model, DFlash, n-gram lookup) can make the same model several times faster, so those runs get their own column. A report that doesn't make clear which kind of run it was counts in neither.
Hardware is grouped by how much it holds and how fast its memory is, since memory speed sets how fast a model writes. Click a model for every report behind its row.