Is qwen/qwen3.8-27b good for coding?

In our benchmarks, qwen/qwen3.8-27b scores 63.9/100 for coding. It runs at about 40.8 tok/s, so if you want more speed, ornith-1.5-35b-a3b-apex-mtp-i-quality is faster (~105.2 tok/s) and still scores well for coding (83.6/100).

Is qwen/qwen3.8-27b good for agentic (tool-using) tasks?

In our benchmarks, qwen/qwen3.8-27b scores 83.5/100 for agentic workflows. It runs at about 40.8 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 qwen/qwen3.8-27b for local inference?

Across 75 community benchmark runs, qwen/qwen3.8-27b reaches up to 156.8 tok/s and averages 40.8 tok/s, with the fastest results on NVIDIA GeForce RTX 5090.

How much memory does qwen/qwen3.8-27b need?

The leanest observed configuration used about 16.3 GB of memory (quantizations tested: 4bit, 6bit, Q2_K_XL, Q4_K_M, Q8_0, nvfp4).

Which tools have been used to run qwen/qwen3.8-27b?

Benchmarks were submitted using LM Studio, Ollama, Strata, llama.cpp, ninfer, radiance, vLLM. Results are community-contributed and updated as new runs arrive.

Model benchmarks

qwen/qwen3.8-27b local LLM performance

As of November 2026, qwen/qwen3.8-27b runs at up to 156.8 tok/s for local inference (best of 75 community benchmark runs across 20 GPUs).

LM StudioOllamaStratallama.cppninferradiancevLLM4bit6bitQ2_K_XLQ4_K_MQ8_0nvfp4

ShareReddit X

Model size

27B

Peak speed

156.8 tok/s

Average speed

40.8 tok/s

Avg PP

726.8 tok/s

Min memory

16.3 GB

Max context

262,000 tokens

Avg output / run

45,100 tokens

Avg runtime / run

23m 59s

Avg quality

78.6

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 75 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	45.0	78.6	90.0
Agent Workflow	64.6	83.5	91.9
Code Generation	6.5	63.9	88.1
Role Play & Narrative	77.3	89.0	95.3
Research & Analysis	3.6	78.1	90.3

Performance by hardware and tool

Every hardware/tool/quantization combination qwen/qwen3.8-27b has been benchmarked on, ranked by peak token generation speed. Last updated November 2026.

Hardware	Tool	Quant	Best tok/s	Avg tok/s	Memory	Context	Quality	Runs
NVIDIA GeForce RTX 5090	ninfer	—	156.8 tok/s	150.0 tok/s	n/a	65,536 tokens	68.0	2
NVIDIA GeForce RTX 5090	vLLM	—	155.2 tok/s	155.2 tok/s	n/a	65,536 tokens	89.1	1
NVIDIA L4	Ollama	Q4_K_M	109.3 tok/s	109.3 tok/s	16.7 GB	65,536 tokens	86.7	1
NVIDIA GeForce RTX 3090	vLLM	—	94.9 tok/s	94.9 tok/s	n/a	8,192 tokens	22.7	1
NVIDIA A10G	ninfer	—	79.4 tok/s	79.4 tok/s	n/a	65,536 tokens	85.0	1
AMD Radeon AI PRO R9700	radiance	—	79.3 tok/s	77.1 tok/s	n/a	65,536 tokens	87.6	4
NVIDIA GeForce RTX 4090	ninfer	—	75.5 tok/s	57.0 tok/s	24.2 GB	262,000 tokens	86.3	2
NVIDIA GeForce RTX 3090	llama.cpp	—	56.6 tok/s	44.4 tok/s	22.1 GB	180,000 tokens	86.7	9
AMD Radeon RX 7900 XTX	LM Studio	—	54.4 tok/s	52.7 tok/s	21.6 GB	65,536 tokens	73.6	3
AMD Radeon RX 7900 XTX	LM Studio	Q4_K_M	54.4 tok/s	51.3 tok/s	23.2 GB	65,536 tokens	71.7	2
NVIDIA GeForce RTX 3090	Strata	—	54.0 tok/s	45.0 tok/s	22.8 GB	200,000 tokens	89.5	4
AMD Radeon RX 7900 XTX	llama.cpp	—	52.4 tok/s	51.8 tok/s	n/a	65,536 tokens	58.9	4
NVIDIA GeForce RTX 5070 Ti	LM Studio	—	46.9 tok/s	42.8 tok/s	n/a	121,072 tokens	87.9	2
AMD Radeon RX 7900 XTX	Ollama	Q4_K_M	39.4 tok/s	35.6 tok/s	16.3 GB	65,536 tokens	82.8	5
NVIDIA GeForce RTX 4090	LM Studio	—	39.0 tok/s	39.0 tok/s	n/a	8,192 tokens	66.7	1
Apple M5 Pro	Ollama	nvfp4	33.8 tok/s	33.8 tok/s	19.7 GB	65,536 tokens	81.7	1
Apple M5 Max	LM Studio	4bit	31.0 tok/s	31.0 tok/s	22.3 GB	65,536 tokens	77.6	1
AMD Radeon RX 7600 XT	LM Studio	Q2_K_XL	26.7 tok/s	26.7 tok/s	16.7 GB	65,536 tokens	90.0	1
Apple M5 Max	LM Studio	—	24.3 tok/s	20.1 tok/s	n/a	123,904 tokens	76.4	4
NVIDIA GeForce RTX 5060 Ti	LM Studio	Q4_K_M	23.1 tok/s	22.8 tok/s	17.8 GB	16,384 tokens	84.5	2
Apple M1 Max	LM Studio	4bit	17.8 tok/s	16.2 tok/s	17.7 GB	65,536 tokens	75.2	8
Advanced Micro Devices, Inc. [AMD/ATI] Device 7550	Ollama	Q4_K_M	17.2 tok/s	17.2 tok/s	17.1 GB	8,192 tokens	86.5	1
NVIDIA GeForce GTX 1080 Ti	LM Studio	Q4_K_M	17.2 tok/s	16.6 tok/s	24.1 GB	65,536 tokens	78.7	2
Apple M5 Pro	LM Studio	—	17.1 tok/s	15.2 tok/s	n/a	65,536 tokens	66.7	2
NVIDIA GeForce RTX 3090	LM Studio	Q4_K_M	16.5 tok/s	16.5 tok/s	26.8 GB	177,840 tokens	86.2	1
Apple M1 Max	LM Studio	Q4_K_M	16.0 tok/s	16.0 tok/s	22.1 GB	32,768 tokens	83.3	1
Apple M1 Max	LM Studio	Q8_0	15.4 tok/s	15.4 tok/s	33.5 GB	32,768 tokens	87.4	1
AMD Radeon RX 9070 XT	Ollama	Q4_K_M	15.0 tok/s	15.0 tok/s	16.9 GB	8,192 tokens	85.0	1
Apple M4 Pro	LM Studio	—	14.0 tok/s	14.0 tok/s	n/a	32,768 tokens	85.7	1
Apple M1 Max	LM Studio	6bit	12.5 tok/s	12.5 tok/s	23.9 GB	8,192 tokens	66.3	1
AMD Radeon RX 9070 XT	LM Studio	Q4_K_M	8.8 tok/s	8.8 tok/s	18.3 GB	8,192 tokens	65.2	1
Apple M5	LM Studio	—	7.6 tok/s	7.6 tok/s	n/a	8,192 tokens	43.1	1
AMD Radeon RX 9060 XT	LM Studio	Q4_K_M	5.8 tok/s	5.8 tok/s	n/a	16,384 tokens	81.3	1
NVIDIA RTX 2000 Ada Generation Laptop GPU	Ollama	Q4_K_M	4.8 tok/s	4.6 tok/s	31.9 GB	65,536 tokens	87.1	2

Benchmark runs

All 75 qwen/qwen3.8-27b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is qwen/qwen3.8-27b good for coding?: In our benchmarks, qwen/qwen3.8-27b scores 63.9/100 for coding. It runs at about 40.8 tok/s, so if you want more speed, ornith-1.5-35b-a3b-apex-mtp-i-quality is faster (~105.2 tok/s) and still scores well for coding (83.6/100).
Is qwen/qwen3.8-27b good for agentic (tool-using) tasks?: In our benchmarks, qwen/qwen3.8-27b scores 83.5/100 for agentic workflows. It runs at about 40.8 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 qwen/qwen3.8-27b for local inference?: Across 75 community benchmark runs, qwen/qwen3.8-27b reaches up to 156.8 tok/s and averages 40.8 tok/s, with the fastest results on NVIDIA GeForce RTX 5090.
How much memory does qwen/qwen3.8-27b need?: The leanest observed configuration used about 16.3 GB of memory (quantizations tested: 4bit, 6bit, Q2_K_XL, Q4_K_M, Q8_0, nvfp4).
Which tools have been used to run qwen/qwen3.8-27b?: Benchmarks were submitted using LM Studio, Ollama, Strata, llama.cpp, ninfer, radiance, vLLM. Results are community-contributed and updated as new runs arrive.