Model benchmarks

ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS local LLM performance

As of September 2026, ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS runs at up to 71.9 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).

llama.cpp
ShareRedditX

Model size

27B

Peak speed

71.9 tok/s

Average speed

71.5 tok/s

Avg PP

603.2 tok/s

Min memory

13.2 GB

Max context

65.536 tokens

Avg output / run

45.035 tokens

Avg runtime / run

9m 51s

Avg quality

83.5

Benchmark runs

2

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 2 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)
Overall79.783.587.4
Agent Workflow82.485.989.4
Code Generation68.074.079.9
Role Play & Narrative88.690.692.6
Research & Analysis79.783.787.8

Performance by hardware and tool

Every hardware/tool/quantization combination ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900Mllama.cpp—71.9 tok/s71.5 tok/s13.2 GB65.536 tokens83.52

Benchmark runs

All 2 ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS good for coding?
In our benchmarks, ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS scores 74.0/100 for coding. It runs at about 71.5 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 ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS good for agentic (tool-using) tasks?
In our benchmarks, ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS scores 85.9/100 for agentic workflows. It runs at about 71.5 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MTP-UD-IQ3_XXS is faster (~103.5 tok/s) and still scores well for agentic workflows (88.3/100).
How fast is ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS for local inference?
Across 2 community benchmark runs, ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS reaches up to 71.9 tok/s and averages 71.5 tok/s, with the fastest results on AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M.
How much memory does ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS need?
The leanest observed configuration used about 13.2 GB of memory.
Which tools have been used to run ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Swift-1.5-Qwen3.8-27B-IQ4_XS?
Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.