Model benchmarks
ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp local LLM performance
As of October 2026, ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp runs at up to 32.2 tok/s for local inference (best of 2 community benchmark runs across 2 GPUs).
Model size
27B
Peak speed
32.2 tok/s
Average speed
21.4 tok/s
Avg PP
181.5 tok/s
Min memory
20.5 GB
Max context
262,144 tokens
Avg output / run
16,249 tokens
Avg runtime / run
16m 13s
Avg quality
81.2
Benchmark runs
2
GPUs tested
2
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 | 79.6 | 81.2 | 82.7 |
| Agent Workflow | 78.4 | 82.6 | 86.9 |
| Code Generation | 66.7 | 70.7 | 74.8 |
| Role Play & Narrative | 80.9 | 85.3 | 89.7 |
| Research & Analysis | 84.6 | 86.1 | 87.6 |
Performance by hardware and tool
Every hardware/tool/quantization combination ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp 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 |
|---|---|---|---|---|---|---|---|---|
| Advanced Micro Devices, Inc. [AMD/ATI] Device 7551 | llama.cpp | — | 32.2 tok/s | 32.2 tok/s | n/a | 65,536 tokens | 79.5 | 1 |
| Apple M4 Pro | llama.cpp (MTP) | — | 10.7 tok/s | 10.7 tok/s | 20.5 GB | 262,144 tokens | 82.9 | 1 |
Benchmark runs
All 2 ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp good for coding?
- In our benchmarks, ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp scores 70.7/100 for coding. It runs at about 21.4 tok/s, so if you want more speed, tensorfold-qwen38-27b is faster (~25.0 tok/s) and still scores well for coding (84.3/100).
- Is ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp good for agentic (tool-using) tasks?
- In our benchmarks, ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp scores 82.6/100 for agentic workflows. It runs at about 21.4 tok/s, so if you want more speed, muse-glimmer:latest is faster (~33.3 tok/s) and still scores well for agentic workflows (91.6/100).
- How fast is ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp for local inference?
- Across 2 community benchmark runs, ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp reaches up to 32.2 tok/s and averages 21.4 tok/s, with the fastest results on Advanced Micro Devices, Inc. [AMD/ATI] Device 7551.
- How much memory does ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp need?
- The leanest observed configuration used about 20.5 GB of memory.
- Which tools have been used to run ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF:IQ3_S-mtp?
- Benchmarks were submitted using llama.cpp, llama.cpp (MTP). Results are community-contributed and updated as new runs arrive.