METALBENCH

COMPARISON DESK

Compare runs without hiding the mismatches.

A faster number means little if the hardware, task set, or concurrency changed.

Select two or more runs
ComparableWarning: Runtime differsWarning: Quant differsWarning: Reasoning effort differs
RunScoreDecodeRuntimeConfigurationEvidence
Qwen3.6 NVFP4qwen36-nvfp4-agent-2026-08-287/961.5 tok/sFreeTokenNVFP4 · 32,768 ctx · c1Documented local result
Flash-Next Q3_K_XLflash-next-q3kxl-agent-2026-08-289/914.5 tok/sllama.cppQ3_K_XL · 32,768 ctx · c1Documented local result