Model benchmarks

qwen3.8:27b-mtp-q4_K_M local LLM performance

As of September 2026, qwen3.8:27b-mtp-q4_K_M runs at up to 42.9 tok/s for local inference (best of 4 community benchmark runs across 2 GPUs).

OllamaQ4_K_M
ShareRedditX

Model size

27.3B

Peak speed

42.9 tok/s

Average speed

36.6 tok/s

Avg PP

417.9 tok/s

Min memory

16.2 GB

Max context

65,536 tokens

Avg output / run

16,343 tokens

Avg runtime / run

7m 24s

Avg quality

84.7

Benchmark runs

4

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 4 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)
Overall80.684.788.0
Agent Workflow86.187.989.9
Code Generation61.774.182.4
Role Play & Narrative83.389.493.3
Research & Analysis84.187.288.9

Performance by hardware and tool

Every hardware/tool/quantization combination qwen3.8:27b-mtp-q4_K_M has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
AMD Radeon RX 7900 XTXOllamaQ4_K_M42.9 tok/s40.8 tok/s16.2 GB65,536 tokens83.53
AMD Radeon RX 9050 / 9060 XTOllamaQ4_K_M24.2 tok/s24.2 tok/s19.1 GB65,536 tokens88.11

Benchmark runs

All 4 qwen3.8:27b-mtp-q4_K_M runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is qwen3.8:27b-mtp-q4_K_M good for coding?
In our benchmarks, qwen3.8:27b-mtp-q4_K_M scores 74.1/100 for coding. It runs at about 36.6 tok/s, so if you want more speed, K2-Horizon-MoVA-36B-A4B-MLX-4bit is faster (~48.5 tok/s) and still scores well for coding (82.5/100).
Is qwen3.8:27b-mtp-q4_K_M good for agentic (tool-using) tasks?
In our benchmarks, qwen3.8:27b-mtp-q4_K_M scores 88.0/100 for agentic workflows. It runs at about 36.6 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-mtp-q4_K_M for local inference?
Across 4 community benchmark runs, qwen3.8:27b-mtp-q4_K_M reaches up to 42.9 tok/s and averages 36.6 tok/s, with the fastest results on AMD Radeon RX 7900 XTX.
How much memory does qwen3.8:27b-mtp-q4_K_M need?
The leanest observed configuration used about 16.2 GB of memory (quantizations tested: Q4_K_M).
Which tools have been used to run qwen3.8:27b-mtp-q4_K_M?
Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.