Model benchmarks
Swift-1.5-Qwen3.8-27B-IQ4_XS local LLM performance
As of September 2026, Swift-1.5-Qwen3.8-27B-IQ4_XS runs at up to 40.5 tok/s for local inference (best of 3 community benchmark runs across 2 GPUs).
Model size
27B
Peak speed
40.5 tok/s
Average speed
39.7 tok/s
Avg PP
518.2 tok/s
Min memory
n/a
Max context
262,144 tokens
Avg output / run
33,570 tokens
Avg runtime / run
13m 28s
Avg quality
82.3
Benchmark runs
3
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 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.
| Task | P5 (low) | Avg | P95 (high) |
|---|---|---|---|
| Overall | 80.1 | 82.3 | 85.6 |
| Agent Workflow | 74.9 | 79.3 | 82.0 |
| Code Generation | 74.0 | 78.1 | 82.6 |
| Role Play & Narrative | 79.3 | 84.6 | 88.7 |
| Research & Analysis | 84.1 | 87.0 | 90.4 |
Performance by hardware and tool
Every hardware/tool/quantization combination Swift-1.5-Qwen3.8-27B-IQ4_XS has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| NVIDIA TITAN RTX | llama.cpp | — | 40.5 tok/s | 40.2 tok/s | n/a | 128,000 tokens | 83.2 | 2 |
| AMD Radeon RX 6700/6700 XT/6750 XT / 6800M/6850M XT | llama.cpp | — | 38.8 tok/s | 38.8 tok/s | n/a | 262,144 tokens | 80.4 | 1 |
Benchmark runs
All 3 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 Swift-1.5-Qwen3.8-27B-IQ4_XS good for coding?
- In our benchmarks, Swift-1.5-Qwen3.8-27B-IQ4_XS scores 78.1/100 for coding. It runs at about 39.7 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 Swift-1.5-Qwen3.8-27B-IQ4_XS good for agentic (tool-using) tasks?
- In our benchmarks, Swift-1.5-Qwen3.8-27B-IQ4_XS scores 79.3/100 for agentic workflows. It runs at about 39.7 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-1.5-Qwen3.8-27B-IQ4_XS for local inference?
- Across 3 community benchmark runs, Swift-1.5-Qwen3.8-27B-IQ4_XS reaches up to 40.5 tok/s and averages 39.7 tok/s, with the fastest results on NVIDIA TITAN RTX.
- Which tools have been used to run Swift-1.5-Qwen3.8-27B-IQ4_XS?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.