Is qwen3-coder:30b good for agentic (tool-using) tasks?

In our benchmarks, qwen3-coder:30b scores 53.5/100 for agentic workflows. It runs at about 116.0 tok/s, so if you want more speed, Nex-N2.5-mini-MLX-4bit is faster (~137.0 tok/s) and still scores well for agentic workflows (81.2/100).

How fast is qwen/qwen3-coder-30b for local inference?

Across 4 community benchmark runs, qwen/qwen3-coder-30b reaches up to 240.2 tok/s and averages 116.0 tok/s, with the fastest results on NVIDIA GeForce RTX 5090.

How much memory does qwen/qwen3-coder-30b need?

The leanest observed configuration used about 18.9 GB of memory (quantizations tested: Q4_K_M).

Which tools have been used to run qwen/qwen3-coder-30b?

Benchmarks were submitted using LM Studio, Ollama. Results are community-contributed and updated as new runs arrive.

Model benchmarks

qwen/qwen3-coder-30b local LLM performance

As of September 2026, qwen/qwen3-coder-30b runs at up to 240.2 tok/s for local inference (best of 4 community benchmark runs across 3 GPUs).

LM StudioOllamaQ4_K_M

ShareReddit X

Model size

30B

Peak speed

240.2 tok/s

Average speed

116.0 tok/s

Avg PP

904.1 tok/s

Min memory

18.9 GB

Max context

65,536 tokens

Avg output / run

7,240 tokens

Avg runtime / run

2m 9s

Avg quality

58.3

Benchmark runs

GPUs tested

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 4 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	54.7	58.3	64.1
Agent Workflow	42.1	53.5	71.6
Code Generation	55.2	58.0	60.5
Role Play & Narrative	61.3	63.5	67.0
Research & Analysis	54.8	58.3	62.8

Performance by hardware and tool

Every hardware/tool/quantization combination qwen/qwen3-coder-30b 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
NVIDIA GeForce RTX 5090	Ollama	Q4_K_M	240.2 tok/s	240.2 tok/s	18.9 GB	32,768 tokens	56.4	1
Apple M5 Pro	LM Studio	—	84.5 tok/s	84.5 tok/s	n/a	32,768 tokens	54.5	1
Apple M5 Pro	Ollama	Q4_K_M	82.9 tok/s	82.9 tok/s	20.3 GB	32,768 tokens	65.4	1
Apple M5 Max	LM Studio	—	56.5 tok/s	56.5 tok/s	n/a	65,536 tokens	57.0	1

Benchmark runs

All 4 qwen/qwen3-coder-30b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is qwen3-coder:30b good for agentic (tool-using) tasks?: In our benchmarks, qwen3-coder:30b scores 53.5/100 for agentic workflows. It runs at about 116.0 tok/s, so if you want more speed, Nex-N2.5-mini-MLX-4bit is faster (~137.0 tok/s) and still scores well for agentic workflows (81.2/100).
How fast is qwen/qwen3-coder-30b for local inference?: Across 4 community benchmark runs, qwen/qwen3-coder-30b reaches up to 240.2 tok/s and averages 116.0 tok/s, with the fastest results on NVIDIA GeForce RTX 5090.
How much memory does qwen/qwen3-coder-30b need?: The leanest observed configuration used about 18.9 GB of memory (quantizations tested: Q4_K_M).
Which tools have been used to run qwen/qwen3-coder-30b?: Benchmarks were submitted using LM Studio, Ollama. Results are community-contributed and updated as new runs arrive.