Best hardware for each model

Pick a model, get the hardware: the smallest card that runs it well, the fastest machine, and whether a Mac does the job. Engine estimates, updated with the dataset (2026-10-13).

Qwen3 8B

8B params: Q4_K_M

From NVIDIA GeForce RTX 4060: runs on Mac Mini M6 16GB

Qwen3 14B

14B params: Q4_K_M

From NVIDIA GeForce RTX 4060 Ti: runs on Mac Mini M6 16GB

GPT-OSS 20B

21B params: MXFP4

From NVIDIA GeForce RTX 4060 Ti: runs on MacBook Air M5 24GB

Qwen3.8 27B

27B params: Q4_K_M

From AMD Radeon RX 7900 XT: runs on MacBook Air M5 24GB

Llama 3.3 70B Instruct

70B params: Q4_K_M

From NVIDIA RTX PRO 6000 Blackwell: runs on MacBook Pro M5 Max 128GB

GPT-OSS 120B

117B params: MXFP4

From NVIDIA RTX PRO 6000 Blackwell: runs on MacBook Pro M5 Max 128GB

Qwen3.5 35B-A3B Instruct

35B params: Q4_K_M

From NVIDIA GeForce RTX 3090: runs on MacBook Pro M5 Pro 48GB

Gemma 4 26B-A4B

26B params: Q4_K_M

From AMD Radeon RX 7900 XT: runs on MacBook Air M5 24GB

Llama 4 Scout

109B params: Q4_K_M

From NVIDIA RTX PRO 6000 Blackwell: runs on MacBook Pro M5 Max 128GB

Nemotron 3.5 Lightning 30B-A3B

30B params: Q4_K_M

From NVIDIA GeForce RTX 5090: runs on MacBook Pro M5 Pro 48GB

tok/s are ModelFit engine estimates (bandwidth-bound), not measurements. Data CC BY 4.0.