Model benchmarks
Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp local LLM performance
As of October 2026, Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp runs at up to 59.1 tok/s for local inference (best of 7 community benchmark runs across 5 GPUs).
Model size
27B
Peak speed
59.1 tok/s
Average speed
34.3 tok/s
Avg PP
448.1 tok/s
Min memory
10.7 GB
Max context
131,072 tokens
Avg output / run
35,429 tokens
Avg runtime / run
19m 31s
Avg quality
83.8
Benchmark runs
7
GPUs tested
5
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 7 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 | 83.8 | 88.4 |
| Agent Workflow | 75.0 | 83.4 | 90.8 |
| Code Generation | 68.1 | 77.4 | 85.1 |
| Role Play & Narrative | 83.5 | 90.1 | 94.6 |
| Research & Analysis | 77.4 | 84.3 | 87.9 |
Performance by hardware and tool
Every hardware/tool/quantization combination Qwen3.8-27B-GSQ-RCO-IQ3_XXS-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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon RX 9070/9070 XT/9070 GRE | llama.cpp | — | 59.1 tok/s | 45.5 tok/s | n/a | 128,144 tokens | 80.1 | 2 |
| NVIDIA GeForce RTX 5060 | llama.cpp | — | 40.0 tok/s | 39.4 tok/s | n/a | 131,072 tokens | 83.7 | 2 |
| AMD Radeon RX 6700/6700 XT/6750 XT / 6800M/6850M XT | llama.cpp | — | 28.6 tok/s | 28.6 tok/s | n/a | 32,768 tokens | 84.5 | 1 |
| AMD Radeon RX 9060 XT | Ollama | IQ3_S | 28.5 tok/s | 28.5 tok/s | 10.7 GB | 65,536 tokens | 89.8 | 1 |
| NVIDIA GeForce RTX 3080 Ti | llama.cpp | — | 13.1 tok/s | 13.1 tok/s | n/a | 65,536 tokens | 84.9 | 1 |
Benchmark runs
All 7 Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp good for coding?
- In our benchmarks, Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp scores 77.4/100 for coding. It runs at about 34.3 tok/s, so if you want more speed, nvidia/Qwen3.8-27B-NVFP4 is faster (~63.3 tok/s) and still scores well for coding (82.7/100).
- Is Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp good for agentic (tool-using) tasks?
- In our benchmarks, Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp scores 83.4/100 for agentic workflows. It runs at about 34.3 tok/s, so if you want more speed, Nex-N2.5-mini-APEX-Mini is faster (~88.2 tok/s) and still scores well for agentic workflows (91.3/100).
- How fast is Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp for local inference?
- Across 7 community benchmark runs, Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp reaches up to 59.1 tok/s and averages 34.3 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_XXS-mtp need?
- The leanest observed configuration used about 10.7 GB of memory (quantizations tested: IQ3_S).
- Which tools have been used to run Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp?
- Benchmarks were submitted using Ollama, llama.cpp. Results are community-contributed and updated as new runs arrive.