Model benchmarks
mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF local LLM performance
As of September 2026, mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF runs at up to 31.8 tok/s for local inference (best of 4 community benchmark runs across 1 GPU).
Model size
35B
Peak speed
31.8 tok/s
Average speed
31.6 tok/s
Min memory
n/a
Max context
262,144 tokens
Avg output / run
6,973 tokens
Avg runtime / run
3m 38s
Avg quality
66.8
Benchmark runs
4
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 4 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 | 64.1 | 66.8 | 69.1 |
| Agent Workflow | 51.7 | 59.9 | 66.2 |
| Code Generation | 62.0 | 65.6 | 69.4 |
| Role Play & Narrative | 65.1 | 68.0 | 72.9 |
| Research & Analysis | 69.3 | 73.8 | 76.4 |
Performance by hardware and tool
Every hardware/tool/quantization combination mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| Advanced Micro Devices, Inc. [AMD/ATI] HawkPoint1 | llama.cpp | — | 31.8 tok/s | 31.6 tok/s | n/a | 262,144 tokens | 66.8 | 4 |
Benchmark runs
All 4 mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF good for coding?
- In our benchmarks, mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF scores 65.6/100 for coding. It runs at about 31.6 tok/s, so if you want more speed, local-qwen is faster (~60.2 tok/s) and still scores well for coding (84.8/100).
- Is mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF good for agentic (tool-using) tasks?
- In our benchmarks, mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF scores 59.9/100 for agentic workflows. It runs at about 31.6 tok/s, so if you want more speed, granite-4.2-8b-Q4_K_L is faster (~71.6 tok/s) and still scores well for agentic workflows (92.9/100).
- How fast is mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF for local inference?
- Across 4 community benchmark runs, mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF reaches up to 31.8 tok/s and averages 31.6 tok/s, with the fastest results on Advanced Micro Devices, Inc. [AMD/ATI] HawkPoint1.
- Which tools have been used to run mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.