The pool, cell by cell
pool snapshot 2026-09-18 · every row is a real community cell with its basis and run count · ranking provisional · a signed run outranks any claim
reading the pool...
pool snapshot 2026-09-18 · every row is a real community cell with its basis and run count · ranking provisional · a signed run outranks any claim
reading the pool...
49 cells shown · 14 measured · 35 reported · MIN_RUNS_MEASURED = 3 · capture or correct a number ->
| # | model | rig | basis | value | n | ctx | bits |
|---|---|---|---|---|---|---|---|
| 1 | LFM2.5-8B-A1B-GGUFchat | RTX 3090 24GB ×2 | measured | 276.3 tok/s | 4 | 128,000 | 8-bit |
| 2 | Qwen3.6-35B-A3B-NVFP4chat | RTX 3090 24GB ×2 | measured | 233.4 tok/s | 11 | 262,144 | 4-bit |
| 3 | NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4chat | RTX 3090 24GB ×2 | reported | 197 tok/s | 2 | 131,072 | 4-bit |
| 4 | Qwen3.8-27B-W4A16-AutoRoundchat | RTX 3090 24GB ×2 | measured | 182.4 tok/s | 3 | 262,144 | 4-bit |
| 5 | diffusiongemma-26B-A4B-it-FP8-dynamicchat | RTX 3090 24GB ×2 | reported | 177.6 tok/s | 1 | 262,144 | 8-bit |
| 6 | Qwen3.6-35B-A3Bchat | RTX 3090 24GB ×2 | measured | 155.6 tok/s | 5 | 32,768 | 4-bit |
| 7 | gemma-4-26B-A4B-itchat | RTX 3090 24GB ×2 | reported | 154.1 tok/s | 2 | 262,144 | 8-bit |
| 8 | Qwen3.6-35B-A3B-FP8chat | RTX 3090 24GB ×2 | reported | 140 tok/s | 1 | 32,768 | 8-bit |
| 9 | Ornstein3.6-35B-A3B-GGUFchat | RTX 3090 24GB ×2 | reported | 138 tok/s | 1 | 8,192 | 4-bit |
| 10 | Qwen3.6-35B-A3Bchat | RTX 3090 24GB ×2 | reported | 137.9 tok/s | 1 | 2,048 | 6-bit |
| 11 | Ornith-1.0-35B-GGUFchat | RTX 3090 24GB ×2 | reported | 136.5 tok/s | 1 | 262,144 | 4-bit |
| 12 | gemma-4-31B-itchat | RTX 3090 24GB ×2 | reported | 135.1 tok/s | 1 | 2,048 | 4-bit |
| 13 | Qwen3.6-35B-A3B-GGUFchat | RTX 3090 24GB ×2 | reported | 125.9 tok/s | 1 | 131,072 | 3-bit |
| 14 | Qwen3.6-35B-A3B-GGUFchat | RTX 3090 24GB ×2 | reported | 124 tok/s | 1 | 1,048,576 | 4-bit |
| 15 | Carnice-9b-W8A16-AWQchat | RTX 3090 24GB ×2 | reported | 114.3 tok/s | 1 | 2,048 | 4-bit |
| 16 | Qwen3.5-9B-GGUFchat | RTX 3090 24GB ×2 | reported | 110.1 tok/s | 1 | 8,192 | 4-bit |
| 17 | Qwen3.6-35B-A3B-NSC-ACE-SABER-GGUFchat | RTX 3090 24GB ×2 | reported | 106.2 tok/s | 1 | 2,048 | 8-bit |
| 18 | Qwen3.8-27B-GPTQ-4bitchat | RTX 3090 24GB ×2 | reported | 96.6 tok/s | 1 | 8,192 | 4-bit |
| 19 | gemma-4-31B-itchat | RTX 3090 24GB ×2 | reported | 87.7 tok/s | 1 | 8,192 | 8-bit |
| 20 | Qwen3.6-27B-int4-AutoRoundchat | RTX 3090 24GB ×2 | measured | 78.2 tok/s | 7 | 262,144 | 4-bit |
| 21 | Qwen3.8-27B-GGUFchat | RTX 3090 24GB ×2 | reported | 73 tok/s | 1 | 2,048 | 4-bit |
| 22 | Qwen3.8-27Bchat | RTX 3090 24GB ×2 | measured | 72.8 tok/s | 3 | 262,144 | 4-bit |
| 23 | Qwen3.5-9Bchat | RTX 3090 24GB ×2 | reported | 71.7 tok/s | 1 | 2,048 | 8-bit |
| 24 | Qwen3.5-9B-MTP-GGUFchat | RTX 3090 24GB ×2 | reported | 70.6 tok/s | 1 | 262,144 | 8-bit |
| 25 | Qwen3.6-27B-MTP-GGUFchat | RTX 3090 24GB ×2 | measured | 65.2 tok/s | 7 | 262,144 | 4-bit |
| 26 | ThinkingCap-Qwen3.6-27Bchat | RTX 3090 24GB ×2 | measured | 58.6 tok/s | 6 | 131,072 | 8-bit |
| 27 | Qwen3.6-27B-AWQ-BF16-INT4-mtp-bf16chat | RTX 3090 24GB ×2 | reported | 57.5 tok/s | 2 | 16,384 | 4-bit |
| 28 | Qwopus3.6-27B-v2-GGUFchat | RTX 3090 24GB ×2 | reported | 56.4 tok/s | 1 | 262,144 | 6-bit |
| 29 | Qwen3.8-Flash-Next-REAP-256-duo-GGUFchat | RTX 3090 24GB ×2 | reported | 53.6 tok/s | 2 | 32,768 | 3-bit |
| 30 | Qwen3.8-27B-GGUFchat | RTX 3090 24GB ×2 | measured | 51.6 tok/s | 5 | 262,144 | 8-bit |
| 31 | Ornstein3.6-27B-MTP-NSC-ACE-SABER-GGUFchat | RTX 3090 24GB ×2 | reported | 49.5 tok/s | 1 | 2,048 | 4-bit |
| 32 | gemma-4-12b-it-GGUFchat | RTX 3090 24GB ×2 | measured | 48.7 tok/s | 3 | 131,072 | 8-bit |
| 33 | Ornstein3.6-27B-MTP-NSC-ACE-SABER-GGUFchat | RTX 3090 24GB ×2 | measured | 47.2 tok/s | 3 | 2,048 | 6-bit |
| 34 | Qwen3.6-27Bchat | RTX 3090 24GB ×2 | reported | 46.7 tok/s | 1 | 2,048 | 4-bit |
| 35 | Qwen3.6-27B-FP8chat | RTX 3090 24GB ×2 | measured | 46.3 tok/s | 4 | 262,144 | 8-bit |
| 36 | Qwen3.5-27B-int4-AutoRoundchat | RTX 3090 24GB ×2 | reported | 45.6 tok/s | 1 | 130,000 | 4-bit |
| 37 | Qwen3.6-27Bchat | RTX 3090 24GB ×2 | reported | 45.5 tok/s | 1 | 8,192 | 8-bit |
| 38 | Laguna-S-2.1-GGUFchat | RTX 3090 24GB ×2 | reported | 43.6 tok/s | 2 | 2,048 | 4-bit |
| 39 | Qwen3.6-27B-AWQ-BF16-INT4chat | RTX 3090 24GB ×2 | reported | 43.2 tok/s | 2 | 2,048 | 4-bit |
| 40 | Qwen3.8-Flash-Next-GGUFchat | RTX 3090 24GB ×2 | reported | 41.6 tok/s | 2 | 217,088 | 3-bit |
| 41 | Qwen3.6-27Bchat | RTX 3090 24GB ×2 | measured | 41.2 tok/s | 4 | 131,712 | 4-bit |
| 42 | Qwen3.6-27Bchat | RTX 3090 24GB ×2 | reported | 40.9 tok/s | 1 | 2,048 | 8-bit |
| 43 | Qwen3.6-35B-A3B-GGUFchat | RTX 3090 24GB ×2 | reported | 40.3 tok/s | 1 | 32,768 | 6-bit |
| 44 | Qwen3.6-27B-GGUFchat | RTX 3090 24GB ×2 | measured | 37.8 tok/s | 3 | 65,536 | 4-bit |
| 45 | Muse-Glimmer-30B-GGUFchat | RTX 3090 24GB ×2 | reported | 31.7 tok/s | 1 | 2,048 | 6-bit |
| 46 | Qwen3.6-27B-GGUFchat | RTX 3090 24GB ×2 | reported | 23 tok/s | 2 | 32,768 | 8-bit |
| 47 | Qwen3.6-27Bchat | RTX 3090 24GB ×2 | reported | 22.7 tok/s | 1 | 2,048 | 6-bit |
| 48 | Qwen3.6-27B-GGUFchat | RTX 3090 24GB ×2 | reported | 22.2 tok/s | 1 | 32,768 | 6-bit |
| 49 | gemma-4-31B-it-GGUFchat | RTX 3090 24GB ×2 | reported | 14.4 tok/s | 1 | 32,768 | 6-bit |
Rows are community cells (medians), ranked provisionally. A signed run outranks any claim.