Model benchmarks
Qwen3.8-27B-oQ8e-mtp local LLM performance
As of November 2026, Qwen3.8-27B-oQ8e-mtp runs at up to 72.5 tok/s for local inference (best of 134 community benchmark runs across 3 GPUs).
Model size
27B
Peak speed
72.5 tok/s
Average speed
37.6 tok/s
Avg PP
421.8 tok/s
Min memory
32.4 GB
Max context
262,144 tokens
Avg output / run
36,720 tokens
Avg runtime / run
16m 29s
Avg quality
82.8
Benchmark runs
134
GPUs tested
3
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 134 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 | 70.0 | 82.8 | 88.6 |
| Agent Workflow | 59.6 | 81.0 | 91.0 |
| Code Generation | 60.8 | 75.8 | 85.5 |
| Role Play & Narrative | 76.4 | 88.7 | 95.4 |
| Research & Analysis | 77.5 | 85.5 | 90.3 |
Performance by hardware and tool
Every hardware/tool/quantization combination Qwen3.8-27B-oQ8e-mtp has been benchmarked on, ranked by peak token generation speed. Last updated November 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| Apple M5 Ultra | oMLX | oQ8e | 72.5 tok/s | 71.9 tok/s | 32.8 GB | 256,000 tokens | 84.8 | 2 |
| Apple M5 Max | oMLX | oQ8e | 61.4 tok/s | 45.3 tok/s | 32.6 GB | 262,144 tokens | 83.1 | 37 |
| Apple M5 Max | oMLX | — | 39.4 tok/s | 33.1 tok/s | 32.4 GB | 262,144 tokens | 81.8 | 29 |
| Apple M4 Max | oMLX | oQ8e | 37.7 tok/s | 34.3 tok/s | 37.1 GB | 262,144 tokens | 82.8 | 29 |
| Apple M4 Max | oMLX | — | 35.2 tok/s | 34.0 tok/s | 49.2 GB | 262,144 tokens | 83.0 | 37 |
Benchmark runs
All 134 Qwen3.8-27B-oQ8e-mtp runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is Qwen3.8-27B-oQ8e-mtp good for coding?
- In our benchmarks, Qwen3.8-27B-oQ8e-mtp scores 75.8/100 for coding. It runs at about 37.6 tok/s, so if you want more speed, turbo is faster (~124.1 tok/s) and still scores well for coding (85.3/100).
- Is Qwen3.8-27B-oQ8e-mtp good for agentic (tool-using) tasks?
- In our benchmarks, Qwen3.8-27B-oQ8e-mtp scores 81.0/100 for agentic workflows. It runs at about 37.6 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 Qwen3.8-27B-oQ8e-mtp for local inference?
- Across 134 community benchmark runs, Qwen3.8-27B-oQ8e-mtp reaches up to 72.5 tok/s and averages 37.6 tok/s, with the fastest results on Apple M5 Ultra.
- How much memory does Qwen3.8-27B-oQ8e-mtp need?
- The leanest observed configuration used about 32.4 GB of memory (quantizations tested: oQ8e).
- Which tools have been used to run Qwen3.8-27B-oQ8e-mtp?
- Benchmarks were submitted using oMLX. Results are community-contributed and updated as new runs arrive.