Model benchmarks

Swift-Qwen3.8-27b-oQ4e-fp16-mtp local LLM performance

As of September 2026, Swift-Qwen3.8-27b-oQ4e-fp16-mtp runs at up to 48.6 tok/s for local inference (best of 3 community benchmark runs across 1 GPU).

oMLXFP16
ShareRedditX

Model size

27B

Peak speed

48.6 tok/s

Average speed

41.5 tok/s

Avg PP

517.5 tok/s

Min memory

19.6 GB

Max context

262.144 tokens

Avg output / run

31.281 tokens

Avg runtime / run

11m 59s

Avg quality

84.7

Benchmark runs

3

GPUs tested

1

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)
Overall82.184.786.6
Agent Workflow82.485.991.4
Code Generation72.177.083.1
Role Play & Narrative85.490.494.5
Research & Analysis83.185.487.7

Performance by hardware and tool

Every hardware/tool/quantization combination Swift-Qwen3.8-27b-oQ4e-fp16-mtp has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
Apple M5 MaxoMLXFP1648.6 tok/s41.5 tok/s19.6 GB262.144 tokens84.73

Benchmark runs

All 3 Swift-Qwen3.8-27b-oQ4e-fp16-mtp runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is Swift-Qwen3.8-27b-oQ4e-fp16-mtp good for coding?
In our benchmarks, Swift-Qwen3.8-27b-oQ4e-fp16-mtp scores 77.0/100 for coding. It runs at about 41.5 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 Swift-Qwen3.8-27b-oQ4e-fp16-mtp good for agentic (tool-using) tasks?
In our benchmarks, Swift-Qwen3.8-27b-oQ4e-fp16-mtp scores 85.9/100 for agentic workflows. It runs at about 41.5 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 Swift-Qwen3.8-27b-oQ4e-fp16-mtp for local inference?
Across 3 community benchmark runs, Swift-Qwen3.8-27b-oQ4e-fp16-mtp reaches up to 48.6 tok/s and averages 41.5 tok/s, with the fastest results on Apple M5 Max.
How much memory does Swift-Qwen3.8-27b-oQ4e-fp16-mtp need?
The leanest observed configuration used about 19.6 GB of memory (quantizations tested: FP16).
Which tools have been used to run Swift-Qwen3.8-27b-oQ4e-fp16-mtp?
Benchmarks were submitted using oMLX. Results are community-contributed and updated as new runs arrive.