MLX Builders · open data
MLX Bench
How fast does each Mac run each model? Tokens per second and peak memory for open models on Apple's MLX framework, measured on real Macs and published for everyone. Part of MLX Builders.
Add your Mac
Two commands in Terminal on any Apple Silicon Mac. It picks a model that fits your memory, measures it with mlx-lm, and shares the numbers here.
Add --model mlx-community/… to test another model, --context 8192 or 65536 for long prompts, --wallet you.mlx to credit the run, or --dry-run to keep it local. Pods of Mac Studios run it under mlx.launch --backend jaccl. The script is short, so read it first: bench.py.
No runs yet
Be the first: run the script above and your Mac becomes the first row of the open MLX leaderboard.