Model benchmarks
qwen3:8b local LLM performance
As of September 2026, qwen3:8b runs at up to 82.4 tok/s for local inference (best of 3 community benchmark runs across 3 GPUs).
Model size
8.2B
Peak speed
82.4 tok/s
Average speed
55.4 tok/s
Avg PP
2533.3 tok/s
Min memory
5.9 GB
Max context
8,192 tokens
Avg output / run
14,900 tokens
Avg runtime / run
6m 12s
Avg quality
40.2
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 | 31.9 | 40.2 | 51.3 |
| Agent Workflow | 65.9 | 70.3 | 74.8 |
| Code Generation | 0.0 | 0.0 | 0.0 |
| Role Play & Narrative | 60.5 | 68.0 | 80.2 |
| Research & Analysis | 1.3 | 22.6 | 50.3 |
Performance by hardware and tool
Every hardware/tool/quantization combination qwen3:8b 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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon RX 9070 XT | Ollama | Q4_K_M | 82.4 tok/s | 82.4 tok/s | 5.9 GB | 8,192 tokens | 31.4 | 1 |
| NVIDIA GeForce RTX 3060 Ti | Ollama | Q4_K_M | 59.5 tok/s | 59.5 tok/s | 6.1 GB | 8,192 tokens | 53.0 | 1 |
| Apple M5 | LM Studio | — | 24.2 tok/s | 24.2 tok/s | n/a | 8,192 tokens | 36.1 | 1 |
Benchmark runs
All 3 qwen3:8b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is qwen3:8b good for coding?
- In our benchmarks, qwen3:8b scores 0.0/100 for coding. It runs at about 55.4 tok/s, so if you want more speed, ornith-ai/Ornith-1.5-35B-A3B-GGUF is faster (~80.3 tok/s) and still scores well for coding (82.1/100).
- Is qwen3:8b good for agentic (tool-using) tasks?
- In our benchmarks, qwen3:8b scores 70.3/100 for agentic workflows. It runs at about 55.4 tok/s, so if you want more speed, swift-iq2_xs is faster (~91.5 tok/s) and still scores well for agentic workflows (91.4/100).
- How fast is qwen3:8b for local inference?
- Across 3 community benchmark runs, qwen3:8b reaches up to 82.4 tok/s and averages 55.4 tok/s, with the fastest results on AMD Radeon RX 9070 XT.
- How much memory does qwen3:8b need?
- The leanest observed configuration used about 5.9 GB of memory (quantizations tested: Q4_K_M).
- Which tools have been used to run qwen3:8b?
- Benchmarks were submitted using LM Studio, Ollama. Results are community-contributed and updated as new runs arrive.