Tiel-Coder-35B achieves 121.4 tok/s for local inference

DerTomsn · reddit · 2026-08-26

A Reddit user shared benchmark results for Tiel-Coder-35B-A3B-MLX-oQ4. The model achieves up to 121.4 tokens per second locally with decent output quality. The MTP version offers 5 more tok/s, pending real-world tests.

Original post →

More from Models

Models channel →