Model benchmarks
Qwen3-8B-Q4_K_M local LLM performance
As of October 2026, Qwen3-8B-Q4_K_M runs at up to 32.7 tok/s for local inference (best of 2 community benchmark runs across 2 GPUs).
Model size
8B
Peak speed
32.7 tok/s
Average speed
22.6 tok/s
Avg PP
150.9 tok/s
Min memory
5.8 GB
Max context
32,768 tokens
Avg output / run
27,569 tokens
Avg runtime / run
18m 21s
Avg quality
49.9
Benchmark runs
2
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 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.
| Task | P5 (low) | Avg | P95 (high) |
|---|---|---|---|
| Overall | 39.4 | 49.9 | 60.4 |
| Agent Workflow | 58.4 | 62.4 | 66.4 |
| Code Generation | 2.5 | 24.6 | 46.7 |
| Role Play & Narrative | 74.7 | 80.1 | 85.4 |
| Research & Analysis | 3.3 | 32.5 | 61.8 |
Performance by hardware and tool
Every hardware/tool/quantization combination Qwen3-8B-Q4_K_M 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 |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 4060 Laptop GPU | llama.cpp | Q4_K_M | 32.7 tok/s | 32.7 tok/s | 7.7 GB | 32,768 tokens | 61.6 | 1 |
| Apple M3 | Ollama | Q4_K_M | 12.6 tok/s | 12.6 tok/s | 5.8 GB | 8,192 tokens | 38.2 | 1 |
Benchmark runs
All 2 Qwen3-8B-Q4_K_M runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is qwen3:8b-q4_K_M good for coding?
- In our benchmarks, qwen3:8b-q4_K_M scores 24.6/100 for coding. It runs at about 22.6 tok/s, so if you want more speed, byteshape/Qwen3.8-27B-GGUF:IQ4_XS-ASCII-Condensed is faster (~42.0 tok/s) and still scores well for coding (86.3/100).
- Is qwen3:8b-q4_K_M good for agentic (tool-using) tasks?
- In our benchmarks, qwen3:8b-q4_K_M scores 62.4/100 for agentic workflows. It runs at about 22.6 tok/s, so if you want more speed, muse-glimmer:latest is faster (~33.3 tok/s) and still scores well for agentic workflows (91.6/100).
- How fast is Qwen3-8B-Q4_K_M for local inference?
- Across 2 community benchmark runs, Qwen3-8B-Q4_K_M reaches up to 32.7 tok/s and averages 22.6 tok/s, with the fastest results on NVIDIA GeForce RTX 4060 Laptop GPU.
- How much memory does Qwen3-8B-Q4_K_M need?
- The leanest observed configuration used about 5.8 GB of memory (quantizations tested: Q4_K_M).
- Which tools have been used to run Qwen3-8B-Q4_K_M?
- Benchmarks were submitted using Ollama, llama.cpp. Results are community-contributed and updated as new runs arrive.