Is Qwen3.8-27B-oQ6e-mtp good for coding?

In our benchmarks, Qwen3.8-27B-oQ6e-mtp scores 76.6/100 for coding. It runs at about 42.6 tok/s, so if you want more speed, local-qwen is faster (~60.2 tok/s) and still scores well for coding (84.8/100).

Is Qwen3.8-27B-oQ6e-mtp good for agentic (tool-using) tasks?

In our benchmarks, Qwen3.8-27B-oQ6e-mtp scores 79.1/100 for agentic workflows. It runs at about 42.6 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-oQ6e-mtp for local inference?

Across 37 community benchmark runs, Qwen3.8-27B-oQ6e-mtp reaches up to 61.5 tok/s and averages 42.6 tok/s, with the fastest results on Apple M5 Max.

How much memory does Qwen3.8-27B-oQ6e-mtp need?

The leanest observed configuration used about 25.4 GB of memory (quantizations tested: oQ6e).

Which tools have been used to run Qwen3.8-27B-oQ6e-mtp?

Benchmarks were submitted using mlx-serve, oMLX. Results are community-contributed and updated as new runs arrive.

Model benchmarks

Qwen3.8-27B-oQ6e-mtp local LLM performance

As of October 2026, Qwen3.8-27B-oQ6e-mtp runs at up to 61.5 tok/s for local inference (best of 37 community benchmark runs across 2 GPUs).

mlx-serveoMLXoQ6e

ShareReddit X

Model size

27B

Peak speed

61.5 tok/s

Average speed

42.6 tok/s

Avg PP

381.7 tok/s

Min memory

25.4 GB

Max context

262,144 tokens

Avg output / run

23,802 tokens

Avg runtime / run

9m 23s

Avg quality

82.9

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 37 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	75.9	82.9	88.0
Agent Workflow	58.9	79.1	91.0
Code Generation	68.4	76.6	85.7
Role Play & Narrative	82.8	89.9	94.8
Research & Analysis	79.5	85.9	90.4

Performance by hardware and tool

Every hardware/tool/quantization combination Qwen3.8-27B-oQ6e-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
Apple M5 Max	mlx-serve	—	61.5 tok/s	60.5 tok/s	26.7 GB	65,536 tokens	78.8	4
Apple M5 Max	oMLX	oQ6e	52.2 tok/s	50.2 tok/s	25.4 GB	262,144 tokens	84.3	6
Apple M4 Max	oMLX	oQ6e	39.8 tok/s	38.6 tok/s	27.8 GB	262,144 tokens	83.1	26
Apple M5 Max	oMLX	—	29.2 tok/s	29.2 tok/s	26.3 GB	262,144 tokens	85.7	1

Benchmark runs

All 37 Qwen3.8-27B-oQ6e-mtp 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-oQ6e-mtp good for coding?: In our benchmarks, Qwen3.8-27B-oQ6e-mtp scores 76.6/100 for coding. It runs at about 42.6 tok/s, so if you want more speed, local-qwen is faster (~60.2 tok/s) and still scores well for coding (84.8/100).
Is Qwen3.8-27B-oQ6e-mtp good for agentic (tool-using) tasks?: In our benchmarks, Qwen3.8-27B-oQ6e-mtp scores 79.1/100 for agentic workflows. It runs at about 42.6 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-oQ6e-mtp for local inference?: Across 37 community benchmark runs, Qwen3.8-27B-oQ6e-mtp reaches up to 61.5 tok/s and averages 42.6 tok/s, with the fastest results on Apple M5 Max.
How much memory does Qwen3.8-27B-oQ6e-mtp need?: The leanest observed configuration used about 25.4 GB of memory (quantizations tested: oQ6e).
Which tools have been used to run Qwen3.8-27B-oQ6e-mtp?: Benchmarks were submitted using mlx-serve, oMLX. Results are community-contributed and updated as new runs arrive.