inference/research
draftlow · 0.40updated 2026-10-15

SiliconScore community row: Apple M3 Ultra, MLX, GLM-5 4bit MLX, decode 16.7 t/s

16.7 tok/s avg decode (single-request)
This record is a draft. Not yet verified. Provisional. Treat every number on this page as provisional.

Fields

id
benchmarks/ss-m3-ultra-glm5-4bit-decode
Metric
decode_tok_s
Value
16.7
Unit
tok/s avg decode (single-request)
Model
GLM-5 (4bit MLX)
Methodology
COMMUNITY-UNVERIFIED row aggregated by siliconscore.com M3 Ultra chip page (accessed 2026-10-15); underlying evidence is the linked r/LocalLLaMA/HF measurement. MLX runtime, 4-bit-ish quant as named. avg decode tok/s single-request. Low-confidence forum-class evidence. RAM tier: 512 GB. Best of five GLM-5 rows on the same machine (prompt tok/s 187.0 on this row).
Measured by
third_party
Reproducible
no

Cross-references