Gemma 4 E2B vs Qwen3 VL 4B (Reasoning)
Wins 2 of 7 areas
Coding · Following instructions
Wins 4 of 7 areas
Agents · Reasoning · Images and charts · Long documents
Qwen3 VL 4B (Reasoning) wins more areas, narrowly.Gemma 4 E2B is better at coding.
Scores updated · 32 tests both models report · How we compare
Where each one wins
Tests won in each of the seven areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- Images and chartsUnderstanding pictures, charts and video05Qwen3 VL 4B (Reasoning)5 of 5 tests
- ReasoningHard problems that need careful thinking01Qwen3 VL 4B (Reasoning)1 of 3 tests · 2 ties
- AgentsCarrying out multi-step tasks on its own01Qwen3 VL 4B (Reasoning)1 of 1 test
- Long documentsFinding answers in very long texts01Qwen3 VL 4B (Reasoning)1 of 1 test
- CodingWriting and fixing software21Gemma 4 E2B2 of 3 tests
- Following instructionsDoing exactly what it is asked10Gemma 4 E2B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
Agents, long documents and following instructions rest on a single test each.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3 VL 4B (Reasoning) pulls ahead
Where Gemma 4 E2B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+59.4points ahead
- Code for real scientific research problemsSciCode+3.8points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+1.5points ahead
Every test, side by side
All 32 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemma 4 E2B
- LiveCodeBench v6Qwen3 VL 4B (Reasoning) by 7.34451.3+7.3
- SciCodeGemma 4 E2B by 3.820.917.1+3.8
- Terminal-Bench HardGemma 4 E2B by 1.531.5+1.5
ReasoningQwen3 VL 4B (Reasoning)
- GPQA DiamondQwen3 VL 4B (Reasoning) by 6.143.349.4+6.1
- Humanity's Last Examtie4.84.6tie
- CritPttie00tie
FactsEven
- AA-Omniscience · Non-hallucinationGemma 4 E2B by 59.467.68.2+59.4
- AA-Omniscience · AccuracyQwen3 VL 4B (Reasoning) by 5.66.612.2+5.6
Images and chartsQwen3 VL 4B (Reasoning)
Long documentsQwen3 VL 4B (Reasoning)
- AA-LCRQwen3 VL 4B (Reasoning) by 516.321.3+5
Following instructionsGemma 4 E2B
- IFBenchGemma 4 E2B by 1.43836.6+1.4
Other results16 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- AA-OmniscienceGemma 4 E2B by 44.8-23.6-68.4+44.8
- ChartQAQwen3 VL 4B (Reasoning) by 39.843.583.3+39.8
- MMMU (val) (Pass@1)Qwen3 VL 4B (Reasoning) by 29.741.170.8+29.7
- OCRBench v2_enQwen3 VL 4B (Reasoning) by 18.343.561.8+18.3
- TextVQA-valQwen3 VL 4B (Reasoning) by 18.162.580.5+18.1
- MMLU-ProQwen3 VL 4B (Reasoning) by 13.66073.6+13.6
- RealWorldQAQwen3 VL 4B (Reasoning) by 13.26073.2+13.2
- AA Agentic IndexQwen3 VL 4B (Reasoning) by 12.91.514.4+12.9
- BFCL v3Qwen3 VL 4B (Reasoning) by 10.956.467.3+10.9
- DocVQA-valQwen3 VL 4B (Reasoning) by 985.794.7+9
- MATH-VisionQwen3 VL 4B (Reasoning) by 7.652.460+7.6
- τ²-Bench Telecom (AA run)Gemma 4 E2B by 5.320.815.5+5.3
- HallusionBenchQwen3 VL 4B (Reasoning) by 4.359.864.1+4.3
- AA Intelligencetie7.87tie
- Artificial Analysis Coding Indextie7.26.7tie
- IFEvaltie8382.6tie
Questions people ask
Which is better, Gemma 4 E2B or Qwen3 VL 4B (Reasoning)?
Qwen3 VL 4B (Reasoning) wins four of the seven areas where both have results: agents, reasoning, images and charts and long documents. Gemma 4 E2B wins coding and following instructions. They are level on facts.
Which is better for coding?
Gemma 4 E2B. It wins 2 of the 3 coding tests both models report; Qwen3 VL 4B (Reasoning) wins 1.
How do you compare the two?
We use the 32 benchmark tests both models have published scores on. The verdict counts the 16 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 16 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.