whatcani.run on an M1 Max with 64 GB RAMM1 Max with 64 GB RAM?

Find the best models and how to run them locally, based on real data from 3,087,360 tokens across 603 trials from 2 people.Find the best models and how to run them locally, based on real data from tokens across trials from people.

Docs

~ $ ▌

Models

Results include trials with 4,096 input tokens and 1,024 output tokens only.

							Actions
Qwen3.5-0.8B-4bit	mlx_lm	213.9 tok/s	2,913.2 tok/s	1,443 ms	2.20 GB	202 runs
Qwen3.5-0.8B-MLX-8bit	mlx_lm	168.6 tok/s	2,202.7 tok/s	1,869 ms	2.56 GB	727 runs
Qwen3-0.6B-4bit	mlx_lm	151.6 tok/s	1,895.0 tok/s	2,282 ms	2.13 GB	301 run
Llama-3.2-1B-Instruct-4bit	mlx_lm	130.0 tok/s	1,611.2 tok/s	2,603 ms	2.13 GB	301 run
Qwen3-1.7B-4bit	mlx_lm	113.5 tok/s	1,053.3 tok/s	3,662 ms	2.58 GB	301 run
Qwen3-0.6B-Q8_0.gguf	llama.cpp	108.9 tok/s	2,757.2 tok/s	1,600 ms	1.26 GB	202 runs	None
Qwen3.5-2B-6bit	mlx_lm	96.7 tok/s	1,095.4 tok/s	3,704 ms	3.14 GB	301 run
Qwen3.5-0.8B-UD-IQ2_XXS.gguf	llama.cpp	91.7 tok/s	3,180.2 tok/s	1,288 ms	0.68 GB	11 run
Qwen3.5-0.8B-Q4_K_M.gguf	llama.cpp	74.7 tok/s	2,288.3 tok/s	1,970 ms	0.89 GB	101 run	None
Qwen3.5-0.8B-Q3_K_M.gguf	llama.cpp	72.1 tok/s	2,228.2 tok/s	1,980 ms	0.83 GB	101 run	None
Llama-3.2-3B-Instruct-4bit	mlx_lm	68.4 tok/s	652.0 tok/s	6.06 sec	3.43 GB	301 run
Qwen3.5-4B-4bit	mlx_lm	59.7 tok/s	401.8 tok/s	10.07 sec	4.64 GB	301 run
Qwen3.5-35B-A3B-4bit	mlx_lm	55.9 tok/s	380.4 tok/s	10.70 sec	21.00 GB	312 runs
gpt-oss-20b-MXFP4-Q8	mlx_lm	44.5 tok/s	352.2 tok/s	12.01 sec	13.00 GB	301 run
Qwen3-4B-Instruct-2507-4bit	mlx_lm	39.9 tok/s	290.2 tok/s	14.29 sec	4.08 GB	301 run
Qwen3.5-9b-MLX-4bit	mlx_lm	30.6 tok/s	198.6 tok/s	20.75 sec	7.08 GB	301 run
qwen3-4b-instruct-2507-q8_0.gguf	llama.cpp	30.4 tok/s	495.6 tok/s	8.85 sec	5.02 GB	301 run	None
Qwen3.5-4B-Q4_K_M.gguf	llama.cpp	24.4 tok/s	386.4 tok/s	10.67 sec	3.35 GB	101 run	None
Qwen3.5-27B-4bit	mlx_lm	7.5 tok/s	38.7 tok/s	106.08 sec	18.00 GB	301 run

		Actions
Qwen3.5-0.8B-4bit	213.9 tok/s
Qwen3.5-0.8B-MLX-8bit	168.6 tok/s
Qwen3-0.6B-4bit	151.6 tok/s
Llama-3.2-1B-Instruct-4bit	130.0 tok/s
Qwen3-1.7B-4bit	113.5 tok/s
Qwen3-0.6B-Q8_0.gguf	108.9 tok/s
Qwen3.5-2B-6bit	96.7 tok/s
Qwen3.5-0.8B-UD-IQ2_XXS.gguf	91.7 tok/s
Qwen3.5-0.8B-Q4_K_M.gguf	74.7 tok/s
Qwen3.5-0.8B-Q3_K_M.gguf	72.1 tok/s
Llama-3.2-3B-Instruct-4bit	68.4 tok/s
Qwen3.5-4B-4bit	59.7 tok/s
Qwen3.5-35B-A3B-4bit	55.9 tok/s
gpt-oss-20b-MXFP4-Q8	44.5 tok/s
Qwen3-4B-Instruct-2507-4bit	39.9 tok/s
Qwen3.5-9b-MLX-4bit	30.6 tok/s
qwen3-4b-instruct-2507-q8_0.gguf	30.4 tok/s
Qwen3.5-4B-Q4_K_M.gguf	24.4 tok/s
Qwen3.5-27B-4bit	7.5 tok/s

							Actions

		Actions