Model benchmarks
dirk-qwen3.8-27b@iq3_xxs local LLM performance
As of September 2026, dirk-qwen3.8-27b@iq3_xxs runs at up to 19.3 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
27B
Peak speed
19.3 tok/s
Average speed
19.0 tok/s
Avg PP
611.2 tok/s
Min memory
10.6 GB
Max context
120.000 tokens
Avg output / run
12.468 tokens
Avg runtime / run
11m 10s
Avg quality
83.6
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 | 82.7 | 83.6 | 84.5 |
| Agent Workflow | 83.4 | 84.7 | 86.0 |
| Code Generation | 71.1 | 73.9 | 76.7 |
| Role Play & Narrative | 87.7 | 88.8 | 89.9 |
| Research & Analysis | 86.0 | 86.9 | 87.8 |
Performance by hardware and tool
Every hardware/tool/quantization combination dirk-qwen3.8-27b@iq3_xxs 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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon RX 9070/9070 XT/9070 GRE | LM Studio | — | 19.3 tok/s | 19.0 tok/s | 10.6 GB | 120.000 tokens | 83.6 | 2 |
Benchmark runs
All 2 dirk-qwen3.8-27b@iq3_xxs runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is dirk-qwen3.8-27b@iq3_xxs good for coding?
- In our benchmarks, dirk-qwen3.8-27b@iq3_xxs scores 73.9/100 for coding. It runs at about 19.0 tok/s, so if you want more speed, /home/polaris/models/Tiel-Coder-35B-A3B-MTP-UD-Q6_K_XL.gguf is faster (~35.6 tok/s) and still scores well for coding (84.9/100).
- Is dirk-qwen3.8-27b@iq3_xxs good for agentic (tool-using) tasks?
- In our benchmarks, dirk-qwen3.8-27b@iq3_xxs scores 84.7/100 for agentic workflows. It runs at about 19.0 tok/s, so if you want more speed, muse-glimmer:latest is faster (~34.5 tok/s) and still scores well for agentic workflows (93.0/100).
- How fast is dirk-qwen3.8-27b@iq3_xxs for local inference?
- Across 2 community benchmark runs, dirk-qwen3.8-27b@iq3_xxs reaches up to 19.3 tok/s and averages 19.0 tok/s, with the fastest results on AMD Radeon RX 9070/9070 XT/9070 GRE.
- How much memory does dirk-qwen3.8-27b@iq3_xxs need?
- The leanest observed configuration used about 10.6 GB of memory.
- Which tools have been used to run dirk-qwen3.8-27b@iq3_xxs?
- Benchmarks were submitted using LM Studio. Results are community-contributed and updated as new runs arrive.