Model benchmarks
Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp local LLM performance
As of September 2026, Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp runs at up to 35.0 tok/s for local inference (best of 2 community benchmark runs across 2 GPUs).
Model size
27B
Peak speed
35.0 tok/s
Average speed
31.1 tok/s
Avg prefill
542.7 tok/s
Min memory
n/a
Max context
131,072 tokens
Avg output / run
14,809 tokens
Avg runtime / run
8m 10s
Avg quality
85.8
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 | 84.7 | 85.8 | 86.9 |
| Agent Workflow | 81.2 | 81.7 | 82.2 |
| Code Generation | 79.4 | 84.2 | 88.9 |
| Role Play & Narrative | 89.8 | 90.7 | 91.6 |
| Research & Analysis | 86.1 | 86.6 | 87.1 |
Performance by hardware and tool
Every hardware/tool/quantization combination Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp 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 | vLLM | — | 35.0 tok/s | 35.0 tok/s | n/a | 131,072 tokens | 87.0 | 1 |
| NVIDIA GeForce RTX 5060 | llama.cpp | — | 27.1 tok/s | 27.1 tok/s | 13.2 GB | 98,304 tokens | 84.5 | 1 |
Frequently asked questions
- Is Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp good for coding?
- In our benchmarks, Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp scores 84.2/100 for coding, among the top 2 for coding on consumer hardware (≤24 GB VRAM). It runs at about 31.1 tok/s, so if you want more speed, unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_XL is faster (~36.1 tok/s) and still scores well for coding (83.2/100).
- Is Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp good for agentic (tool-using) tasks?
- In our benchmarks, Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp scores 81.7/100 for agentic workflows. It runs at about 31.1 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-GSQ-RCO-IQ3_S-mtp for local inference?
- Across 2 community benchmark runs, Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp reaches up to 35.0 tok/s and averages 31.1 tok/s, with the fastest results on AMD Radeon RX 9070/9070 XT/9070 GRE.
- How much memory does Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp need?
- The leanest observed configuration used about n/a of memory.
- Which tools have been used to run Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp?
- Benchmarks were submitted using llama.cpp, vLLM. Results are community-contributed and updated as new runs arrive.