Google
Google
Google/Gemma 4 31B IT

Runs

View all benchmark runs for this model family.

	Quant
M4 Pro	ud-4bit	mlx_lm0.31.2	9.4 tok/s	92.3 tok/s	27.00 GB 42%
M4 Pro	ud-4bit	mlx_lm0.31.2	5.5 tok/s	82.0 tok/s	23.00 GB 36%
M4 Max	q4_k_m	llama.cppb8680	10.9 tok/s	112.7 tok/s	13.92 GB 11%
M4 Max	ud-4bit	mlx_lm0.31.2	14.3 tok/s	157.6 tok/s	26.00 GB 72%
M4 Max	q3_k_m	llama.cppb8680	13.6 tok/s	153.4 tok/s	15.43 GB 43%
M4 Max	q4_k_m	llama.cppb8680	16.9 tok/s	159.2 tok/s	19.17 GB 53%
M4 Max	q5_k_m	llama.cppb8680	13.0 tok/s	147.6 tok/s	19.72 GB 55%
GeForce RTX 40906×	q4_k_m	llama.cppb8680	2.4 tok/s	689.0 tok/s	18.54 GB 7%
RTX PRO 4000 Blackwell	q4_k_m	llama.cppb8672	23.5 tok/s	658.3 tok/s	17.42 GB 28%
M4 Max	8bit	mlx_lm0.31.2	12.5 tok/s	154.7 tok/s	37.00 GB 29%