Model benchmarks
m5-qwen3-4b-thinking local LLM performance
As of September 2026, m5-qwen3-4b-thinking runs at up to 89.0 tok/s for local inference (best of 2 community benchmark runs across 0 GPUs).
Model size
4B
Peak speed
89.0 tok/s
Average speed
88.5 tok/s
Avg PP
2358.8 tok/s
Min memory
n/a
Max context
8.192 tokens
Avg output / run
16.084 tokens
Avg runtime / run
3m 18s
Avg quality
48.7
Benchmark runs
2
GPUs tested
0
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 | 45.9 | 48.7 | 51.4 |
| Agent Workflow | 65.0 | 71.6 | 78.2 |
| Code Generation | 0.2 | 1.5 | 2.8 |
| Role Play & Narrative | 69.7 | 74.8 | 79.9 |
| Research & Analysis | 33.1 | 46.9 | 60.7 |
Performance by hardware and tool
Every hardware/tool/quantization combination m5-qwen3-4b-thinking 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 |
|---|---|---|---|---|---|---|---|---|
| CPU only | cliproxyapi | — | 89.0 tok/s | 88.5 tok/s | n/a | 8.192 tokens | 48.7 | 2 |
Benchmark runs
All 2 m5-qwen3-4b-thinking runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is m5-qwen3-4b-thinking good for coding?
- In our benchmarks, m5-qwen3-4b-thinking scores 1.5/100 for coding. It runs at about 88.5 tok/s, so if you want more speed, unsloth/Qwen3.8-27B-GGUF:IQ3_S is faster (~104.3 tok/s) and still scores well for coding (83.0/100).
- Is m5-qwen3-4b-thinking good for agentic (tool-using) tasks?
- In our benchmarks, m5-qwen3-4b-thinking scores 71.6/100 for agentic workflows. It runs at about 88.5 tok/s, so if you want more speed, Ornith-1.5-35B-A3B-BigBang-oQ8e-mtp is faster (~102.7 tok/s) and still scores well for agentic workflows (91.1/100).
- How fast is m5-qwen3-4b-thinking for local inference?
- Across 2 community benchmark runs, m5-qwen3-4b-thinking reaches up to 89.0 tok/s and averages 88.5 tok/s.
- How much memory does m5-qwen3-4b-thinking need?
- The leanest observed configuration used about n/a of memory.
- Which tools have been used to run m5-qwen3-4b-thinking?
- Benchmarks were submitted using cliproxyapi. Results are community-contributed and updated as new runs arrive.