Compare / Qwen3 4B Thinking 2507 vs Gemma 3 4B

Qwen3 4B Thinking 2507 vs Gemma 3 4B

Updated November 2026

Verdict

Qwen3 4B Thinking 2507 scores higher on the Small LM Index: 59.8 vs 19.4. On the 8 benchmarks both labs report, Qwen3 4B Thinking 2507 wins 8 and Gemma 3 4B wins 0. Both need about 4.2 GB of memory at 4-bit quantization. Only Gemma 3 4B has a hosted API on OpenRouter ($0.063 per million tokens); the other has to be self-hosted.

Pick Qwen3 4B Thinking 2507 if…

you want the strongest model overall
you mostly write code
you need math and reasoning
you need precise instruction following
you build agents that call tools
you work with long documents
you need a permissive license (Apache-2.0)

Pick Gemma 3 4B if…

you want a hosted API rather than running it yourself
you need image input

Specifications

	Qwen3 4B Thinking 2507	Gemma 3 4B
Small LM Index	59.8 (likely 57–62)	19.4 (likely 18–21)
Lab	Alibaba (Qwen) · China	Google · United States
Released	2025-08-06	2025-03-12
Parameters	4B (dense)	4.3B (dense)
Memory at 4-bit	~4.2 GB	~3.5 GB
Context window	262K tokens	128K tokens
Input	text	text, image
Reasoning	Always thinks	No thinking
License	Apache-2.0	Gemma
API price / 1M tokens	Self-host only	$0.063 ($0.050 in, $0.10 out)
Hugging Face downloads (30 days)	437.1K	985.4K

Benchmarks

Higher is better. Bold marks the winner where both labs report a score. ≈ marks the index model's estimate for a score that wasn't reported, ✓ an independent measurement and * a score measured by another lab.

Benchmark	Qwen3 4B Thinking 2507	Gemma 3 4B
MMLU-Pro Knowledge	74.3^✓	41.7^✓
GPQA Diamond Knowledge	66.7^✓	29.1^✓
Humanity's Last Exam Reasoning	6.2^✓	5.3^✓
AIME Math	82.7^✓	6.3^✓
LiveCodeBench Coding	64.1^✓	11.2^✓
τ²-Bench Agentic	25.4^✓	5.0^✓
IFBench Instruction following	49.8^✓	28.3^✓
MATH-500 Math	—	76.6^✓
AA-LCR Long context	37.3^✓	6.7^✓

Scores are mostly reported by each lab with its own settings, so small gaps are noise. See the methodology. Open in the interactive comparison to add more models.

Questions

Which is better, Qwen3 4B Thinking 2507 or Gemma 3 4B?

Qwen3 4B Thinking 2507 scores higher on the Small LM Index: 59.8 vs 19.4. On the 8 benchmarks both labs report, Qwen3 4B Thinking 2507 wins 8 and Gemma 3 4B wins 0.

Which is cheaper to run?

Only Gemma 3 4B has a hosted API on OpenRouter ($0.063 per million tokens); the other has to be self-hosted. Both need about 4.2 GB of memory at 4-bit quantization.

Which has the longer context window?

Qwen3 4B Thinking 2507 has the longer context window: 262K tokens against 128K.

Can I use Qwen3 4B Thinking 2507 and Gemma 3 4B commercially?

Qwen3 4B Thinking 2507 is released under Apache-2.0 and Gemma 3 4B under Gemma. Apache 2.0 and MIT allow commercial use; other licenses can add conditions, so read the license before shipping a product.