Qwen3 VL 30B A3B Instruct vs Qwen3 VL 4B (Reasoning)
Wins 4 of 7 areas
Coding · Reasoning · Images and charts · Long documents
Wins 2 of 7 areas
Agents · Following instructions
Qwen3 VL 30B A3B Instruct wins more areas, narrowly.Qwen3 VL 4B (Reasoning) is better at agents.
Scores updated · 52 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 video42Qwen3 VL 30B A3B Instruct4 of 7 tests · 1 tie
- ReasoningHard problems that need careful thinking20Qwen3 VL 30B A3B Instruct2 of 3 tests · 1 tie
- CodingWriting and fixing software21Qwen3 VL 30B A3B Instruct2 of 3 tests
- Long documentsFinding answers in very long texts10Qwen3 VL 30B A3B Instruct1 of 1 test
- AgentsCarrying out multi-step tasks on its own02Qwen3 VL 4B (Reasoning)2 of 2 tests
- Following instructionsDoing exactly what it is asked01Qwen3 VL 4B (Reasoning)1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
Long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Qwen3 VL 30B A3B Instruct pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+20.1points ahead
- Questions about short and long videosVideo-MME+14.8points ahead
- Code for real scientific research problemsSciCode+13.7points ahead
Where Qwen3 VL 4B (Reasoning) pulls ahead
- Recent programming contest problemsLiveCodeBench v6+8.7points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+3.5points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+1.9points ahead
Every test, side by side
All 52 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3 VL 30B A3B Instruct
- SciCodeQwen3 VL 30B A3B Instruct by 13.730.817.1+13.7
- LiveCodeBench v6Qwen3 VL 4B (Reasoning) by 8.742.651.3+8.7
- Terminal-Bench HardQwen3 VL 30B A3B Instruct by 4.66.11.5+4.6
AgentsQwen3 VL 4B (Reasoning)
- GDPValQwen3 VL 4B (Reasoning) by 13.8013.8+13.8
- OSWorld-VerifiedQwen3 VL 4B (Reasoning) by 1.130.331.4+1.1
ReasoningQwen3 VL 30B A3B Instruct
- GPQA DiamondQwen3 VL 30B A3B Instruct by 20.169.549.4+20.1
- Humanity's Last ExamQwen3 VL 30B A3B Instruct by 1.76.34.6+1.7
- CritPttie00tie
FactsEven
- AA-Omniscience · AccuracyQwen3 VL 30B A3B Instruct by 3.415.612.2+3.4
- AA-Omniscience · Non-hallucinationQwen3 VL 4B (Reasoning) by 1.96.38.2+1.9
Images and chartsQwen3 VL 30B A3B Instruct
- Video-MMEQwen3 VL 30B A3B Instruct by 14.874.559.7+14.8
- MMMU-ProQwen3 VL 30B A3B Instruct by 10.162.152+10.1
- OCRBenchQwen3 VL 30B A3B Instruct by 9.590.380.8+9.5
- BLINKQwen3 VL 30B A3B Instruct by 4.367.763.4+4.3
- CharXiv (RQ)Qwen3 VL 4B (Reasoning) by 1.448.950.3+1.4
- MMStarQwen3 VL 4B (Reasoning) by 1.172.173.2+1.1
- MathVistatie80.179.5tie
Long documentsQwen3 VL 30B A3B Instruct
- AA-LCRQwen3 VL 30B A3B Instruct by 627.321.3+6
Following instructionsQwen3 VL 4B (Reasoning)
- IFBenchQwen3 VL 4B (Reasoning) by 3.533.136.6+3.5
Other results33 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.
- Arena-Hard-V2Qwen3 VL 30B A3B Instruct by 21.758.536.8+21.7
- ScreenSpot-Pro (No tools)Qwen3 VL 30B A3B Instruct by 11.360.549.2+11.3
- LVBenchQwen3 VL 30B A3B Instruct by 962.553.5+9
- Artificial Analysis Coding IndexQwen3 VL 30B A3B Instruct by 7.614.36.7+7.6
- IncludeQwen3 VL 30B A3B Instruct by 771.664.6+7
- CC-OCRQwen3 VL 30B A3B Instruct by 6.980.773.8+6.9
- SuperGPQAQwen3 VL 30B A3B Instruct by 6.353.146.8+6.3
- MMLU-ProXQwen3 VL 30B A3B Instruct by 5.970.965+5.9
- AIME 2025Qwen3 VL 4B (Reasoning) by 5.269.374.5+5.2
- AA-OmniscienceQwen3 VL 30B A3B Instruct by 5-63.4-68.4+5
- AA Agentic IndexQwen3 VL 4B (Reasoning) by 4.99.514.4+4.9
- CharadesSTAQwen3 VL 30B A3B Instruct by 4.563.559+4.5
- ERQAQwen3 VL 4B (Reasoning) by 4.34347.3+4.3
- MMLU-ProQwen3 VL 30B A3B Instruct by 4.277.873.6+4.2
- τ²-Bench Telecom (AA run)Qwen3 VL 30B A3B Instruct by 3.51915.5+3.5
- MMMU (val) (Pass@1)Qwen3 VL 30B A3B Instruct by 3.474.270.8+3.4
- IFEvalQwen3 VL 30B A3B Instruct by 3.285.882.6+3.2
- MV-BenchQwen3 VL 30B A3B Instruct by 372.369.3+3
- HallusionBenchQwen3 VL 4B (Reasoning) by 2.661.564.1+2.6
- HMMT 2025Qwen3 VL 4B (Reasoning) by 2.550.653.1+2.5
- MMLU-ReduxQwen3 VL 30B A3B Instruct by 2.488.486+2.4
- OCRBench v2 (Chinese)Qwen3 VL 30B A3B Instruct by 257.855.8+2
- ScreenSpotQwen3 VL 30B A3B Instruct by 1.894.792.9+1.8
- OCRBench v2_enQwen3 VL 30B A3B Instruct by 1.463.261.8+1.4
- BFCL v3Qwen3 VL 4B (Reasoning) by 166.367.3+1
- InfoVQA_testQwen3 VL 4B (Reasoning) by 18283+1
- AA Intelligencetie7.97tie
- DocVQA_testtie9594.2tie
- VideoMMMUtie68.769.4tie
- RealWorldQAtie73.773.2tie
- MM MTBenchtie8.17.7tie
- MATH-Visiontie60.260tie
- AI2Dtie8584.9tie
Questions people ask
Which is better, Qwen3 VL 30B A3B Instruct or Qwen3 VL 4B (Reasoning)?
Qwen3 VL 30B A3B Instruct wins four of the seven areas where both have results: coding, reasoning, images and charts and long documents. Qwen3 VL 4B (Reasoning) wins agents and following instructions. They are level on facts.
Which is better for coding?
Qwen3 VL 30B A3B Instruct. 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 52 benchmark tests both models have published scores on. The verdict counts the 19 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 33 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.