Runs

Most recent benchmark runs across all devices and models.


Bonsai 27B Source	M4 Max	llama.cppb9960	41.5 tok/s	205.6 tok/s	5.05 GB 14%	Verified
Qwen3-Coder-30B-A3B-Instruct Source	M4 Pro	mlx_lm0.31.3	64.5 tok/s	776.0 tok/s	20.00 GB 42%	Verified
Qwen2.5-Coder-7B-Instruct Source	M4 Pro	mlx_lm0.31.3	48.2 tok/s	487.9 tok/s	7.97 GB 17%	Verified
Qwen3.6-35B-A3B Source	M4 Pro	mlx_lm0.31.3	51.5 tok/s	873.3 tok/s	37.00 GB 77%	Verified
Qwen3.6-35B-A3B Source	M4 Pro	mlx_lm0.31.3	83.4 tok/s	944.4 tok/s	21.00 GB 44%	Verified
Gemma 4 12B IT Source	M5 Pro	llama.cppb8920	32.9 tok/s	619.7 tok/s	6.05 GB 9%	Verified
Qwen3.5-27B (GLM-5.1 Distill) v1 Source	M5 Pro	llama.cppb8920	19.7 tok/s	282.0 tok/s	9.29 GB 15%	Verified
Qwen3.5-4B Source	M4 Pro	mlx_lm0.31.3	72.3 tok/s	619.9 tok/s	4.68 GB 10%	Verified
Qwen3.6-35B-A3B Source	M4 Pro	mlx_lm0.31.3	73.9 tok/s	752.0 tok/s	21.00 GB 44%	Verified
Qwen3.5-0.8B Source	M4 Pro	mlx_lm0.31.3	195.5 tok/s	4,357.1 tok/s	2.68 GB 6%	Verified


Bonsai 27B Source	41.5 tok/s	Verified
Qwen3-Coder-30B-A3B-Instruct Source	64.5 tok/s	Verified
Qwen2.5-Coder-7B-Instruct Source	48.2 tok/s	Verified
Qwen3.6-35B-A3B Source	51.5 tok/s	Verified
Qwen3.6-35B-A3B Source	83.4 tok/s	Verified
Gemma 4 12B IT Source	32.9 tok/s	Verified
Qwen3.5-27B (GLM-5.1 Distill) v1 Source	19.7 tok/s	Verified
Qwen3.5-4B Source	72.3 tok/s	Verified
Qwen3.6-35B-A3B Source	73.9 tok/s	Verified
Qwen3.5-0.8B Source	195.5 tok/s	Verified