Model benchmarks
Dirk-Qwen3.8-27B-UD-IQ4_XS local LLM performance
As of September 2026, Dirk-Qwen3.8-27B-UD-IQ4_XS runs at up to 38.2 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
27B
Peak speed
38.2 tok/s
Average speed
37.6 tok/s
Avg prefill
508.7 tok/s
Min memory
13.2 GB
Max context
65,536 tokens
Avg output / run
14,731 tokens
Avg runtime / run
6m 43s
Avg quality
87.4
Benchmark runs
2
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 2 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 | 87.1 | 87.4 | 87.7 |
| Agent Workflow | 85.9 | 86.3 | 86.6 |
| Code Generation | 81.9 | 82.4 | 82.9 |
| Role Play & Narrative | 89.7 | 90.9 | 92.2 |
| Research & Analysis | 89.3 | 90.1 | 90.9 |
Performance by hardware and tool
Every hardware/tool/quantization combination Dirk-Qwen3.8-27B-UD-IQ4_XS 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 |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5060 | llama.cpp | — | 38.2 tok/s | 37.6 tok/s | 13.2 GB | 65,536 tokens | 87.4 | 2 |
Frequently asked questions
- Is Dirk-Qwen3.8-27B-UD-IQ4_XS good for coding?
- In our benchmarks, Dirk-Qwen3.8-27B-UD-IQ4_XS scores 82.4/100 for coding. It runs at about 37.6 tok/s, so if you want more speed, K2-Horizon-MoVA-36B-A4B-MLX-4bit is faster (~48.8 tok/s) and still scores well for coding (85.3/100).
- Is Dirk-Qwen3.8-27B-UD-IQ4_XS good for agentic (tool-using) tasks?
- In our benchmarks, Dirk-Qwen3.8-27B-UD-IQ4_XS scores 86.3/100 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 Dirk-Qwen3.8-27B-UD-IQ4_XS for local inference?
- Across 2 community benchmark runs, Dirk-Qwen3.8-27B-UD-IQ4_XS reaches up to 38.2 tok/s and averages 37.6 tok/s, with the fastest results on NVIDIA GeForce RTX 5060.
- How much memory does Dirk-Qwen3.8-27B-UD-IQ4_XS need?
- The leanest observed configuration used about 13.2 GB of memory.
- Which tools have been used to run Dirk-Qwen3.8-27B-UD-IQ4_XS?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.