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.
More from Models
- Together Ranks Top Open Models: Kimi K3 and DeepSeek V4 Lead Use Cases — togethercompute · 2026-08-26
- Questions over Astra's progress: 2 months for 3 more models? — teortaxesTex · 2026-08-26
- View: Tokens-per-second matters more than model size now — natesiggard · 2026-08-26
- Benchmark: Tool Calling Performance of Qwen 35B-A3B Variants — OsmanthusBloom · 2026-08-26
- Grok 4.6 now available on OpenCode Go subscription — veggie_eric · 2026-08-26
- One Claude Design Task Burned 95% of the $100/Mo Plan — zeeg · 2026-08-26