Is Qwen3.8-27B-UD-Q3_K_XL good for coding?

In our benchmarks, Qwen3.8-27B-UD-Q3_K_XL scores 54.2/100 for coding. It runs at about 56.5 tok/s, so if you want more speed, Ornith-1.5-35B-A3B.APEX-I-NanoPlus is faster (~110.5 tok/s) and still scores well for coding (81.1/100).

Is Qwen3.8-27B-UD-Q3_K_XL good for agentic (tool-using) tasks?

In our benchmarks, Qwen3.8-27B-UD-Q3_K_XL scores 84.6/100 for agentic workflows. It runs at about 56.5 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-UD-Q3_K_XL for local inference?

Across 30 community benchmark runs, Qwen3.8-27B-UD-Q3_K_XL reaches up to 78.4 tok/s and averages 56.5 tok/s, with the fastest results on NVIDIA GeForce RTX 5070 Ti.

How much memory does Qwen3.8-27B-UD-Q3_K_XL need?

The leanest observed configuration used about 14.7 GB of memory (quantizations tested: Q3_K).

Which tools have been used to run Qwen3.8-27B-UD-Q3_K_XL?

Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.

Model benchmarks

Qwen3.8-27B-UD-Q3_K_XL local LLM performance

As of October 2026, Qwen3.8-27B-UD-Q3_K_XL runs at up to 78.4 tok/s for local inference (best of 30 community benchmark runs across 3 GPUs).

llama.cppQ3_K

ShareReddit X

Model size

27B

Peak speed

78.4 tok/s

Average speed

56.5 tok/s

Avg PP

607.3 tok/s

Min memory

14.7 GB

Max context

84,992 tokens

Avg output / run

61,103 tokens

Avg runtime / run

18m 38s

Avg quality

76.8

Benchmark runs

GPUs tested

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 30 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	45.4	76.8	88.3
Agent Workflow	74.7	84.6	91.0
Code Generation	0.0	54.2	86.9
Role Play & Narrative	82.9	88.0	95.1
Research & Analysis	28.8	80.2	91.0

Performance by hardware and tool

Every hardware/tool/quantization combination Qwen3.8-27B-UD-Q3_K_XL 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
NVIDIA GeForce RTX 5070 Ti	llama.cpp	Q3_K	78.4 tok/s	61.4 tok/s	15.6 GB	84,992 tokens	76.4	22
NVIDIA GeForce RTX 5080	llama.cpp	Q3_K	55.1 tok/s	52.3 tok/s	14.7 GB	65,536 tokens	85.1	5
NVIDIA GeForce RTX 5060	llama.cpp	Q3_K	37.1 tok/s	36.3 tok/s	n/a	84,992 tokens	83.5	2
CPU only	llama.cpp	Q3_K	9.7 tok/s	9.7 tok/s	n/a	8,192 tokens	28.5	1

Benchmark runs

All 30 Qwen3.8-27B-UD-Q3_K_XL runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is Qwen3.8-27B-UD-Q3_K_XL good for coding?: In our benchmarks, Qwen3.8-27B-UD-Q3_K_XL scores 54.2/100 for coding. It runs at about 56.5 tok/s, so if you want more speed, Ornith-1.5-35B-A3B.APEX-I-NanoPlus is faster (~110.5 tok/s) and still scores well for coding (81.1/100).
Is Qwen3.8-27B-UD-Q3_K_XL good for agentic (tool-using) tasks?: In our benchmarks, Qwen3.8-27B-UD-Q3_K_XL scores 84.6/100 for agentic workflows. It runs at about 56.5 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-UD-Q3_K_XL for local inference?: Across 30 community benchmark runs, Qwen3.8-27B-UD-Q3_K_XL reaches up to 78.4 tok/s and averages 56.5 tok/s, with the fastest results on NVIDIA GeForce RTX 5070 Ti.
How much memory does Qwen3.8-27B-UD-Q3_K_XL need?: The leanest observed configuration used about 14.7 GB of memory (quantizations tested: Q3_K).
Which tools have been used to run Qwen3.8-27B-UD-Q3_K_XL?: Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.