Model benchmarks
Qwen/Qwen2.5-3B-Instruct local LLM performance
As of October 2026, Qwen/Qwen2.5-3B-Instruct runs at up to 74.3 tok/s for local inference (best of 2 community benchmark runs across 2 GPUs).
Model size
3B
Peak speed
74.3 tok/s
Average speed
59.0 tok/s
Avg PP
3654.7 tok/s
Min memory
n/a
Max context
32,768 tokens
Avg output / run
3,875 tokens
Avg runtime / run
1m 27s
Avg quality
35.7
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 | 30.9 | 35.7 | 40.4 |
| Agent Workflow | 41.2 | 54.7 | 68.1 |
| Code Generation | 18.6 | 22.2 | 25.7 |
| Role Play & Narrative | 34.4 | 35.1 | 35.8 |
| Research & Analysis | 29.5 | 30.8 | 32.0 |
Performance by hardware and tool
Every hardware/tool/quantization combination Qwen/Qwen2.5-3B-Instruct 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 7800 XT | vLLM | — | 74.3 tok/s | 74.3 tok/s | n/a | 32,768 tokens | 40.9 | 1 |
| NVIDIA GeForce RTX 3060 | vLLM | — | 43.7 tok/s | 43.7 tok/s | n/a | 8,192 tokens | 30.4 | 1 |
Benchmark runs
All 2 Qwen/Qwen2.5-3B-Instruct runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is Qwen/Qwen2.5-3B-Instruct good for coding?
- In our benchmarks, Qwen/Qwen2.5-3B-Instruct scores 22.2/100 for coding. It runs at about 59.0 tok/s, so if you want more speed, ornith-ai/Ornith-1.5-35B-A3B-GGUF is faster (~90.8 tok/s) and still scores well for coding (83.8/100).
- Is Qwen/Qwen2.5-3B-Instruct good for agentic (tool-using) tasks?
- In our benchmarks, Qwen/Qwen2.5-3B-Instruct scores 54.7/100 for agentic workflows. It runs at about 59.0 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 Qwen/Qwen2.5-3B-Instruct for local inference?
- Across 2 community benchmark runs, Qwen/Qwen2.5-3B-Instruct reaches up to 74.3 tok/s and averages 59.0 tok/s, with the fastest results on AMD Radeon RX 7800 XT.
- Which tools have been used to run Qwen/Qwen2.5-3B-Instruct?
- Benchmarks were submitted using vLLM. Results are community-contributed and updated as new runs arrive.