v4.3
Know your AI.Grow your business.
The open-weights hardware leaderboard. Every system below runs the same model, precision and workload on Metrum AI Bench, so the ranking compares hardware and nothing else.
Open-weights hardware leaderboard
Verified Metrum runs. Each ranking holds one model, one precision and one workload, and every value is measured, never interpolated.
Higher is better. Each system is ranked at its best measured point that holds at least 50 tokens/s per request.
Loading the leaderboard.