llamaperf

GLM-5.3 753B (40B active)

on M5 Ultra 512GB · oMLX · 131,072 ctx

Tone: mixedApple hardware
Nov 16, 2026
Throughput
25.7 t/s gen
Quant
4bit (MLX)
System RAM
512 GB

Summary

User reports GLM-5.3 at 25.7 tok/s on a 1K prompt on a Mac Studio M5 Ultra 512GB, with 23.7 tok/s at 32K and 21.0 tok/s at 128K. Setup is oMLX 0.7.3 with custom kernels, the mlx-community 4-bit MLX weights, 131072 context, MTP off, one user, median of 5 runs. Peak memory was 423.1 GiB. Without the custom kernels a 32K prompt read 2.2x slower and peaked 37.0 GiB higher. User calls it decent but thinks DeepSeek V4.1 Flash or GLM Flash 5.3 would suit the machine better.

Compare this run

Medians of plain runs: one device, one request, no speculative decoding, the whole model in the device's memory. The report above is one setup, so read it beside them and not as one of them.

GLM-5.3 753B · 40B active at 4-bit by hardware

Bars are medians on one scale, 0 to 25.7 tokens/s. Press ▶ to watch text arrive at a row's speed.

Models most run on the M5 Ultra 512GB

Bars are medians on one scale, 0 to 129.2 tokens/s. Press ▶ to watch text arrive at a row's speed.

Everything reported on the M5 Ultra 512GB →Every GLM-5.3 report →Estimate GLM-5.3 on this card →