Every memory option compared, 120 GB/s memory bandwidth. Updated 2 November 2026.
M4 is sold with 16, 24, 32 GB of memory. Counting the 31 models tracked here: 16 GB runs 16 models well; 24 GB runs 20 models well; 32 GB runs 23 models well. Memory bandwidth is 120 GB/s, about the same memory speed as a base M4 (120 GB/s). Apple Silicon shares one pool of memory between the processor and the GPU, and macOS lets the GPU use about two thirds of it on Macs up to 36 GB and about three quarters above that, so the memory you buy decides which models load at all.
| Model | 16 GB | 24 GB | 32 GB |
|---|---|---|---|
| Wan 2.2 14B | No | No | Slow about 53 min |
| LTX-2.3 | No | Slow about 8 min | Good about 4 min |
| Qwen 3.6 35B A3B | No | No | Tight 17/s |
| Qwen 3.5 35B A3B | No | No | Tight 17/s |
| Gemma 4 31B | No | Slow 3/s | Slow 3/s |
| DeepSeek R1 32B | No | Slow 3/s | Slow 3/s |
| Qwen3 Coder 30B | No | Tight 16/s | Great 17/s |
| Qwen 3.6 27B | No | Slow 3/s | Slow 4/s |
| Gemma 4 26B A4B | No | Tight 12/s | Great 14/s |
| Z-Image Turbo 6B | Slow about 2 min | Good about 49 s | Good about 49 s |
| Wan 2.2 5B | Slow about 27 min | Good about 13 min | Good about 13 min |
| gpt-oss 20B | Tight 13/s | Great 14/s | Great 14/s |
| ACE-Step 1.5 | Good about 3 min | Good about 3 min | Good about 3 min |
| FLUX.2 klein 4B | Good about 34 s | Good about 34 s | Good about 34 s |
| Phi-4 14B | Good 7/s | Good 7/s | Good 7/s |
| Gemma 4 12B | Good 8/s | Good 8/s | Good 8/s |
| Stable Diffusion XL | Good about 18 s | Good about 18 s | Good about 18 s |
| Qwen3 TTS 1.7B | Good about 52 s | Good about 52 s | Good about 52 s |
| Gemma 4 E4B | Great 20/s | Great 20/s | Great 20/s |
| Qwen 3.5 9B | Good 10/s | Good 10/s | Good 10/s |
| Qwen3 VL 8B | Good 11/s | Good 11/s | Good 11/s |
| DeepSeek R1 8B | Great 12/s | Great 12/s | Great 12/s |
| Qwen2.5 Coder 7B | Great 14/s | Great 14/s | Great 14/s |
| Gemma 4 E2B | Great 32/s | Great 32/s | Great 32/s |
| Whisper large-v3 turbo | Good about 1 min | Good about 1 min | Good about 1 min |
| Qwen 3.5 4B | Great 19/s | Great 19/s | Great 19/s |
| Kokoro TTS | Good about 11 s | Good about 11 s | Good about 11 s |
Text models show words a second. Image, video and voice models show the time per job (image, 4 second clip, minute of speech or similar).
As an Amazon Associate I earn from qualifying purchases.
About 15 GB is usable by the GPU. The largest chat or coding model that fits is gpt-oss 20B (15.2 GB), generating about 13 words a second with no spare room. The quickest is Gemma 4 E2B at about 32 words a second. Closest miss: Gemma 4 26B A4B, which needs about 19.4 GB (4 GB more than this configuration can give the GPU).
Find M4 Macs with 16 GB on Amazon
About 23 GB is usable by the GPU. The largest chat or coding model that fits is Qwen3 Coder 30B (20.4 GB), generating about 16 words a second with 2.4 GB to spare. The quickest is Gemma 4 E2B at about 32 words a second. Closest miss: Qwen 3.5 35B A3B, which needs about 23.6 GB (1 GB more than this configuration can give the GPU).
Find M4 Macs with 24 GB on Amazon
About 30 GB is usable by the GPU. The largest chat or coding model that fits is Qwen 3.6 35B A3B (24.6 GB), generating about 17 words a second with 5.8 GB to spare. The quickest is Gemma 4 E2B at about 32 words a second. Closest miss: Llama 3.3 70B, which needs about 45.6 GB (15 GB more than this configuration can give the GPU).
For everyday chat, 16 GB runs 16 of 31 models well. 24 GB to 32 GB unlocks 3 more: Qwen 3.5 35B A3B, Qwen 3.6 35B A3B, LTX-2.3.
No. It needs about 21.5 GB of usable memory and this chip's largest option is 32 GB.
16 GB runs gpt-oss 20B at about 13 words a second; 24 GB runs gpt-oss 20B at about 14 words a second; 32 GB runs gpt-oss 20B at about 14 words a second.
As an Amazon Associate I earn from qualifying purchases.
Unlocks 4 more: Gemma 4 31B, Qwen 3.6 27B, DeepSeek R1 32B, Wan 2.2 14B.
Unlocks 7 more: Gemma 4 31B, Llama 3.3 70B, Qwen 3.5 122B A10B, gpt-oss 120B and more.