Model benchmarks
swift-qwen3.8-27b local LLM performance
As of October 2026, swift-qwen3.8-27b runs at up to 49.6 tok/s for local inference (best of 3 community benchmark runs across 3 GPUs).
Model size
27B
Peak speed
49.6 tok/s
Average speed
29.5 tok/s
Avg PP
502.8 tok/s
Min memory
29.8 GB
Max context
80,000 tokens
Avg output / run
25,567 tokens
Avg runtime / run
23m 20s
Avg quality
68.2
Benchmark runs
3
GPUs tested
3
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 | 55.6 | 68.2 | 80.3 |
| Agent Workflow | 70.4 | 81.6 | 91.7 |
| Code Generation | 0.0 | 24.1 | 65.0 |
| Role Play & Narrative | 84.7 | 89.6 | 92.9 |
| Research & Analysis | 57.7 | 77.4 | 90.5 |
Performance by hardware and tool
Every hardware/tool/quantization combination swift-qwen3.8-27b 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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon RX 9050 / 9060 XT | LM Studio | — | 49.6 tok/s | 49.6 tok/s | n/a | 8,192 tokens | 54.1 | 1 |
| AMD Radeon RX 7900 XTX | LM Studio | Q4_K_S | 29.8 tok/s | 29.8 tok/s | 29.8 GB | 80,000 tokens | 81.5 | 1 |
| NVIDIA GeForce RTX 5060 Ti | LM Studio | — | 9.1 tok/s | 9.1 tok/s | n/a | 8,192 tokens | 68.8 | 1 |
Benchmark runs
All 3 swift-qwen3.8-27b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is swift-qwen3.8-27b good for coding?
- In our benchmarks, swift-qwen3.8-27b scores 24.1/100 for coding. It runs at about 29.5 tok/s, so if you want more speed, Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF-Q6_K is faster (~34.2 tok/s) and still scores well for coding (85.9/100).
- Is swift-qwen3.8-27b good for agentic (tool-using) tasks?
- In our benchmarks, swift-qwen3.8-27b scores 81.6/100 for agentic workflows. It runs at about 29.5 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 for local inference?
- Across 3 community benchmark runs, swift-qwen3.8-27b reaches up to 49.6 tok/s and averages 29.5 tok/s, with the fastest results on AMD Radeon RX 9050 / 9060 XT.
- How much memory does swift-qwen3.8-27b need?
- The leanest observed configuration used about 29.8 GB of memory (quantizations tested: Q4_K_S).
- Which tools have been used to run swift-qwen3.8-27b?
- Benchmarks were submitted using LM Studio. Results are community-contributed and updated as new runs arrive.