Model benchmarks

Qwen3.8-27B-UD-IQ3_XXS local LLM performance

As of September 2026, Qwen3.8-27B-UD-IQ3_XXS runs at up to 81.6 tok/s for local inference (best of 3 community benchmark runs across 2 GPUs).

llama.cpp
ShareRedditX

Model size

27B

Peak speed

81.6 tok/s

Average speed

74.5 tok/s

Avg PP

684.4 tok/s

Min memory

13.2 GB

Max context

65.536 tokens

Avg output / run

21.184 tokens

Avg runtime / run

4m 49s

Avg quality

57.8

Benchmark runs

3

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 3 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.

TaskP5 (low)AvgP95 (high)
Overall41.857.874.8
Agent Workflow49.060.068.3
Code Generation0.022.259.9
Role Play & Narrative92.592.893.2
Research & Analysis25.656.278.5

Performance by hardware and tool

Every hardware/tool/quantization combination Qwen3.8-27B-UD-IQ3_XXS has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
NVIDIA GeForce RTX 4080 SUPERllama.cpp81.6 tok/s81.6 tok/s13.2 GB65.536 tokens76.91
NVIDIA GeForce RTX 5070 Tillama.cpp72.1 tok/s70.9 tok/s13.2 GB8.192 tokens48.32

Benchmark runs

All 3 Qwen3.8-27B-UD-IQ3_XXS runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is Qwen3.8-27B-UD-IQ3_XXS good for coding?
In our benchmarks, Qwen3.8-27B-UD-IQ3_XXS scores 22.2/100 for coding. It runs at about 74.5 tok/s, so if you want more speed, unsloth/Qwen3.8-27B-GGUF:IQ3_S is faster (~104.3 tok/s) and still scores well for coding (83.0/100).
Is Qwen3.8-27B-UD-IQ3_XXS good for agentic (tool-using) tasks?
In our benchmarks, Qwen3.8-27B-UD-IQ3_XXS scores 60.0/100 for agentic workflows. It runs at about 74.5 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-Q4_K_S is faster (~165.0 tok/s) and still scores well for agentic workflows (89.3/100).
How fast is Qwen3.8-27B-UD-IQ3_XXS for local inference?
Across 3 community benchmark runs, Qwen3.8-27B-UD-IQ3_XXS reaches up to 81.6 tok/s and averages 74.5 tok/s, with the fastest results on NVIDIA GeForce RTX 4080 SUPER.
How much memory does Qwen3.8-27B-UD-IQ3_XXS need?
The leanest observed configuration used about 13.2 GB of memory.
Which tools have been used to run Qwen3.8-27B-UD-IQ3_XXS?
Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.