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.
Mac
A Mac counts when its report names the product, or when only that product was sold with its chip and memory.
Model size
Apple Silicon 48-96 GB· Apple only
70B-class models fit at 4 bits. Memory speed 273 to 1,229 GB/s across the 18 devices with reports, so a typical figure here spans all of them. 197 single-device reports.
| # | Model | Reports | Gen t/s | PP t/s | With speculation |
|---|---|---|---|---|---|
| 1 | Qwen3.827B Alibaba · mostly Q4_K_M Gen 14 t/s (8 runs) · PP 126 t/s (5 runs) · with speculation 34 t/s (16) | 32 | 14 8 runs, 5 devices | 126 5 runs | 34 16 runs |
| 2 | Qwen3.8 Flash-Next125B · 6B active Alibaba · mostly oQ4e Gen 42 t/s (3 runs) · PP 621 t/s (3 runs) · with speculation 67 t/s (4) | 24 | 42 3 runs, 2 devices | 621 3 runs | 67 4 runs |
| 3 | Qwen3.635B · 3B active Alibaba · mostly 4bit Gen 76 t/s (7 runs) · PP 787 t/s (5 runs) · with speculation 121 t/s (5) | 14 | 76 7 runs, 4 devices | 787 5 runs | 121 5 runs |
| 4 | Qwen3.627B Alibaba · mostly Q4_K_M Gen 13 t/s (4 runs) · PP 124 t/s (3 runs) · with speculation 23 t/s (4) | 9 | 13 4 runs, 3 devices | 124 3 runs | 23 4 runs |
| 5 | Gemma 426B · 4B active Google DeepMind · mostly UD-Q4_K_M Gen 72 t/s (6 runs) · PP 731 t/s (5 runs) | 8 | 72 6 runs, 5 devices | 731 5 runs | none |
| 6 | DeepSeek V4 Flash284B · 13B active DeepSeek · mostly IQ3-XXS Gen 13 t/s (1 run) | 8 | 13 1 run | no PP data | none |
| 7 | Qwen3.59B Alibaba · mostly Q4_K_M Gen 66 t/s (3 runs) · PP 321 t/s (3 runs) · with speculation 129 t/s (2) | 7 | 66 3 runs, 3 devices | 321 3 runs | 129 2 runs |
| 8 | Qwen3.535B · 3B active Alibaba · mostly 4-bit Gen 89 t/s (5 runs) · PP 1789 t/s (4 runs) | 5 | 89 5 runs, 4 devices | 1789 4 runs | none |
| 9 | gpt-oss21B · 3.6B active OpenAI · mostly MXFP4 Gen 78 t/s (4 runs) · PP 768 t/s (4 runs) | 5 | 78 4 runs, 2 devices | 768 4 runs | none |
| 10 | Qwen330B · 3B active Alibaba · mostly Q4_K_M Gen 95 t/s (3 runs) · PP 806 t/s (3 runs) | 4 | 95 3 runs, 2 devices | 806 3 runs | none |
| 11 | Qwen3.54B Alibaba · mostly Q4_K_M Gen 85 t/s (3 runs) · PP 4100 t/s (1 run) | 4 | 85 3 runs, 3 devices | 4100 1 run | none |
| 12 | Gemma 431B Google DeepMind · mostly Q4_K_M Gen 24 t/s (2 runs) · PP 345 t/s (2 runs) | 4 | 24 2 runs, 2 devices | 345 2 runs | none |
| 13 | Qwen38B Alibaba · mostly 4-bit Gen 98 t/s (3 runs) · PP 2948 t/s (2 runs) | 3 | 98 3 runs, 2 devices | 2948 2 runs | none |
| 14 | Gemma 4 E2B5.1B Google DeepMind · mostly Q8_0 Gen 87 t/s (3 runs) · PP 1559 t/s (3 runs) | 3 | 87 3 runs, 2 devices | 1559 3 runs | none |
| 15 | Gemma 312B Google · mostly 4-bit Gen 64 t/s (2 runs) · PP 1778 t/s (2 runs) | 3 | 64 2 runs | 1778 2 runs | none |
| 16 | Gemma 4 E4B8B Google DeepMind · mostly Q8_0 Gen 46 t/s (3 runs) · PP 787 t/s (3 runs) | 3 | 46 3 runs, 2 devices | 787 3 runs | none |
| 17 | Qwen 2.5 Coder32B mostly 4-bit Gen 26 t/s (3 runs) · PP 685 t/s (3 runs) | 3 | 26 3 runs, 2 devices | 685 3 runs | none |
| 18 | Qwen3.527B Alibaba · mostly 4-bit (source config) Gen 25 t/s (2 runs) · PP 461 t/s (3 runs) | 3 | 25 2 runs, 2 devices | 461 3 runs | none |
| 19 | Qwen30.6B Alibaba · mostly 4-bit Gen 514 t/s (2 runs) · PP 22991 t/s (2 runs) | 2 | 514 2 runs | 22991 2 runs | none |
| 20 | Llama 3.21B Meta · mostly 4-bit Gen 440 t/s (1 run) · PP 16968 t/s (1 run) | 2 | 440 1 run | 16968 1 run | none |
| 21 | Qwen31.7B Alibaba · mostly 4-bit Gen 327 t/s (2 runs) · PP 11928 t/s (2 runs) | 2 | 327 2 runs | 11928 2 runs | none |
| 22 | Gemma 31B Google · mostly 4-bit Gen 295 t/s (1 run) · PP 16750 t/s (1 run) | 2 | 295 1 run | 16750 1 run | none |
| 23 | Llama 3.23B Meta · mostly 4-bit Gen 200 t/s (1 run) · PP 6123 t/s (1 run) | 2 | 200 1 run | 6123 1 run | none |
| 24 | Qwen34B Alibaba · mostly 4-bit Gen 173 t/s (2 runs) · PP 5239 t/s (2 runs) | 2 | 173 2 runs | 5239 2 runs | none |
| 25 | Gemma 34B Google · mostly 4-bit Gen 163 t/s (2 runs) · PP 5656 t/s (2 runs) | 2 | 163 2 runs | 5656 2 runs | none |
| 26 | Qwen 2.5 Coder7B mostly 4-bit Gen 114 t/s (2 runs) · PP 3132 t/s (2 runs) | 2 | 114 2 runs | 3132 2 runs | none |
| 27 | Qwen2.5-Omni3B mostly Q8_0 Gen 69 t/s (2 runs) · PP 1202 t/s (2 runs) | 2 | 69 2 runs | 1202 2 runs | none |
| 28 | Qwen314B Alibaba · mostly 4-bit Gen 60 t/s (2 runs) · PP 1661 t/s (2 runs) | 2 | 60 2 runs | 1661 2 runs | none |
| 29 | Qwen 2.5 Coder14B mostly 4-bit Gen 58 t/s (2 runs) · PP 1633 t/s (2 runs) | 2 | 58 2 runs | 1633 2 runs | none |
| 30 | Qwen3-VL4B mostly Q8_0 Gen 54 t/s (2 runs) · PP 918 t/s (2 runs) | 2 | 54 2 runs | 918 2 runs | none |
| 31 | Gemma 412B Google DeepMind · mostly Q4_K_M Gen 30 t/s (2 runs) · PP 292 t/s (2 runs) | 2 | 30 2 runs | 292 2 runs | none |
| 32 | Gemma 327B Google · mostly 4-bit Gen 30 t/s (1 run) · PP 751 t/s (1 run) | 2 | 30 1 run | 751 1 run | none |
| 33 | Qwen332B Alibaba · mostly 4-bit Gen 27 t/s (2 runs) · PP 699 t/s (2 runs) | 2 | 27 2 runs | 699 2 runs | none |
| 34 | Mistral Small 3.124B mostly Q4_K_M Gen 18 t/s (2 runs) · PP 145 t/s (2 runs) | 2 | 18 2 runs | 145 2 runs | none |
| 35 | Kimi K32800B · 104B active Moonshot AI · mostly VQ3R No generation data | 2 | no generation data | no PP data | none |
| 36 | Qwen3.8 Swift-1.527B Alibaba · mostly 4.7bpw No generation data · with speculation 35 t/s (1) | 2 | no generation data | no PP data | 35 1 run |
| 37 | Qwen3.8 Uncensored27B Alibaba · mostly 4-bit No generation data · with speculation 59 t/s (2) | 2 | no generation data | no PP data | 59 2 runs |
| 38 | Ministral 33B mostly 4-bit (source config) Gen 128 t/s (1 run) · PP 743 t/s (1 run) | 1 | 128 1 run | 743 1 run | none |
| 39 | LFM224B · 2B active mostly 4-bit mixed (config) Gen 118 t/s (1 run) · PP 616 t/s (1 run) | 1 | 118 1 run | 616 1 run | none |
| 40 | Qwen3 30B-A3B, standard 8 experts30B · 3.3B active Alibaba · mostly MLX 4-bit affine, group 64 Gen 98 t/s (1 run) | 1 | 98 1 run | no PP data | none |
| 41 | Qwen2.57B Alibaba · mostly 4-bit (source config) Gen 77 t/s (1 run) · PP 603 t/s (1 run) | 1 | 77 1 run | 603 1 run | none |
| 42 | gpt-oss117B · 5.1B active OpenAI · mostly MXFP4 Gen 73 t/s (1 run) · PP 347 t/s (1 run) | 1 | 73 1 run | 347 1 run | none |
| 43 | Phi-3.5 mini3.8B Microsoft · mostly 4bit Gen 72 t/s (1 run) | 1 | 72 1 run | no PP data | none |
| 44 | DeepSeek V4 Flash 9B distill9B DeepSeek · mostly Q4_K_M Gen 44 t/s (1 run) | 1 | 44 1 run | no PP data | none |
| 45 | Qwen3.8 27B dense27B Alibaba · mostly Q4_K_M Gen 21 t/s (1 run) · PP 144 t/s (1 run) | 1 | 21 1 run | 144 1 run | none |
| 46 | Bonsai 2 Ternary-Bonsai-2-27B27B PrismML · mostly PQ2_0 Gen 21 t/s (1 run) · PP 127 t/s (1 run) | 1 | 21 1 run | 127 1 run | none |
| 47 | Muse Glimmer30B Meta · mostly 8-bit Gen 8.2 t/s (1 run) | 1 | 8.2 1 run | no PP data | none |
| 48 | DeepSeek V4.1 Flash552B · 16B active No generation data | 1 | no generation data | no PP data | none |
| 49 | Qwen3-Coder30B · 3B active mostly Q4_K_M No generation data | 1 | no generation data | no PP data | none |
| 50 | Qwen3-Coder-Next80B · 3B active mostly 6-bit No generation data | 1 | no generation data | no PP data | none |
| 51 | Qwen3-Next80B · 3B active Alibaba · mostly 4-bit No generation data | 1 | no generation data | no PP data | none |
| 52 | Qwen3.8 Qwen3.8-27B27B Alibaba · mostly Q4_K_M No generation data | 1 | no generation data | no PP data | none |
| 53 | Qwen3.8 Splash-HQ27B Alibaba · mostly Q8 No generation data · with speculation 37 t/s (1) | 1 | no generation data | no PP data | 37 1 run |
| 54 | Qwen3.8 Swift-1.5 Flash-Next125B · 6B active Alibaba · mostly IQ3_S No generation data · with speculation 145 t/s (1) | 1 | no generation data | no PP data | 145 1 run |
Not in the table: 70 reports don't say which model size ran.
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.