bestmodel.run / pool — agent view Honesty ladder: measured > reported > extrapolated > formula > no data yet. Top 60 cells, default sort: fastest decode. Columns: rig | model | basis | value | n. Filter: rig=max-128gb. Full machine surface: /llms.txt · REST API: api.bestmodel.run rig | model | basis | value | n max-128gb | prism-ml-ternary-bonsai-8b-mlx-2bit | reported | 135 tok/s | n=1 max-128gb | mlx-community-gemma-4-26b-a4b-it-4bit | reported | 117 tok/s | n=1 max-128gb | mlx-community-qwen3-6-35b-a3b-4bit | reported | 109 tok/s | n=1 max-128gb | dealignai-qwen3-6-27b-mxfp4-crack | reported | 33 tok/s | n=1 max-128gb | gestaltlabs-ornstein-3-6-27b-gguf | reported | 22 tok/s | n=1 max-128gb | gestaltlabs-ornstein-hermes-3-6-27b-mlx-8bit | reported | 18 tok/s | n=1 max-128gb | mlx-community-deepseek-v4-flash-2bit-dq | reported | 17 tok/s | n=1 max-128gb | unsloth-qwen3-6-27b-ud-mlx-6bit | reported | 13 tok/s | n=1 Capture or correct a number: /console (signed, Ed25519).