Model benchmarks
Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit local LLM performance
As of October 2026, Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit runs at up to 119.7 tok/s for local inference (best of 3 community benchmark runs across 1 GPU).
Model size
35B
Peak speed
119.7 tok/s
Average speed
111.6 tok/s
Avg PP
890.4 tok/s
Min memory
19.7 GB
Max context
131,072 tokens
Avg output / run
7,084 tokens
Avg runtime / run
1m 18s
Avg quality
71.2
Benchmark runs
3
GPUs tested
1
Quality by task
Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 3 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.
| Task | P5 (low) | Avg | P95 (high) |
|---|---|---|---|
| Overall | 67.3 | 71.2 | 75.5 |
| Agent Workflow | 63.5 | 68.7 | 73.1 |
| Code Generation | 63.6 | 66.2 | 67.9 |
| Role Play & Narrative | 74.5 | 79.5 | 82.9 |
| Research & Analysis | 52.1 | 70.6 | 83.3 |
Performance by hardware and tool
Every hardware/tool/quantization combination Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| Apple M4 Pro | oMLX | 4bit | 119.7 tok/s | 111.6 tok/s | 19.7 GB | 131,072 tokens | 71.2 | 3 |
Benchmark runs
All 3 Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit good for agentic (tool-using) tasks?
- In our benchmarks, Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit scores 68.7/100 for agentic workflows. It runs at about 111.6 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MLX-oQ4e-MTP is faster (~151.5 tok/s) and still scores well for agentic workflows (82.8/100).
- How fast is Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit for local inference?
- Across 3 community benchmark runs, Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit reaches up to 119.7 tok/s and averages 111.6 tok/s, with the fastest results on Apple M4 Pro.
- How much memory does Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit need?
- The leanest observed configuration used about 19.7 GB of memory (quantizations tested: 4bit).
- Which tools have been used to run Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bit?
- Benchmarks were submitted using oMLX. Results are community-contributed and updated as new runs arrive.