Qwen3 VL 30B A3B Instruct vs Qwen3 VL 8B Instruct
Wins 4 of 7 areas
Coding · Reasoning · Images and charts · Long documents
Wins 2 of 7 areas
Agents · Facts
Qwen3 VL 30B A3B Instruct wins more areas, narrowly.Qwen3 VL 8B Instruct is cheaper and better at agents.
Scores updated · 51 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 video51Qwen3 VL 30B A3B Instruct5 of 7 tests · 1 tie
- CodingWriting and fixing software30Qwen3 VL 30B A3B Instruct3 of 3 tests
- ReasoningHard problems that need careful thinking20Qwen3 VL 30B A3B Instruct2 of 3 tests · 1 tie
- Long documentsFinding answers in very long texts10Qwen3 VL 30B A3B Instruct1 of 1 test
- AgentsCarrying out multi-step tasks on its own02Qwen3 VL 8B Instruct2 of 2 tests
- FactsGetting facts right instead of making them up02Qwen3 VL 8B Instruct2 of 2 tests
- Following instructionsDoing exactly what it is asked00Even0 each · 1 tie
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.
Qwen3 VL 8B Instruct costs 12% less for the same work.
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+26.8points ahead
- Harder college exam questions with imagesMMMU-Pro+14.8points ahead
- Code for real scientific research problemsSciCode+13.4points ahead
Where Qwen3 VL 8B Instruct pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+4.9points ahead
- Completes tasks by operating a computer desktopOSWorld-Verified+3.6points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+2.7points ahead
Every test, side by side
All 51 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 13.430.817.4+13.4
- Terminal-Bench HardQwen3 VL 30B A3B Instruct by 3.86.12.3+3.8
- LiveCodeBench v6Qwen3 VL 30B A3B Instruct by 3.342.639.3+3.3
AgentsQwen3 VL 8B Instruct
- GDPValQwen3 VL 8B Instruct by 9.109.1+9.1
- OSWorld-VerifiedQwen3 VL 8B Instruct by 3.630.333.9+3.6
ReasoningQwen3 VL 30B A3B Instruct
- GPQA DiamondQwen3 VL 30B A3B Instruct by 26.869.542.7+26.8
- Humanity's Last ExamQwen3 VL 30B A3B Instruct by 3.66.32.7+3.6
- CritPttie00tie
FactsQwen3 VL 8B Instruct
- AA-Omniscience · AccuracyQwen3 VL 8B Instruct by 4.915.620.5+4.9
- AA-Omniscience · Non-hallucinationQwen3 VL 8B Instruct by 2.76.39+2.7
Images and chartsQwen3 VL 30B A3B Instruct
- MMMU-ProQwen3 VL 30B A3B Instruct by 14.862.147.3+14.8
- Video-MMEQwen3 VL 30B A3B Instruct by 3.174.571.4+3.1
- MathVistaQwen3 VL 30B A3B Instruct by 2.980.177.2+2.9
- CharXiv (RQ)Qwen3 VL 30B A3B Instruct by 2.548.946.4+2.5
- BLINKQwen3 VL 8B Instruct by 1.467.769.1+1.4
- MMStarQwen3 VL 30B A3B Instruct by 1.272.170.9+1.2
- OCRBenchtie90.389.6tie
Long documentsQwen3 VL 30B A3B Instruct
- AA-LCRQwen3 VL 30B A3B Instruct by 10.627.316.7+10.6
Other results32 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.
- AIME 2025Qwen3 VL 30B A3B Instruct by 23.469.345.9+23.4
- HMMT 2025Qwen3 VL 30B A3B Instruct by 18.150.632.5+18.1
- AA-OmniscienceQwen3 VL 8B Instruct by 11.5-63.4-51.9+11.5
- τ²-Bench Telecom (AA run)Qwen3 VL 8B Instruct by 10.21929.2+10.2
- AA Agentic IndexQwen3 VL 8B Instruct by 8.99.518.4+8.9
- SuperGPQAQwen3 VL 30B A3B Instruct by 8.653.144.5+8.6
- CharadesSTAQwen3 VL 30B A3B Instruct by 7.563.556+7.5
- Artificial Analysis Coding IndexQwen3 VL 30B A3B Instruct by 714.37.3+7
- MATH-VisionQwen3 VL 30B A3B Instruct by 6.360.253.9+6.3
- MMLU-ProQwen3 VL 30B A3B Instruct by 6.277.871.6+6.2
- ScreenSpot-Pro (No tools)Qwen3 VL 30B A3B Instruct by 5.960.554.6+5.9
- MMLU-ProXQwen3 VL 30B A3B Instruct by 5.570.965.4+5.5
- IncludeQwen3 VL 30B A3B Instruct by 4.671.667+4.6
- MMMU (val) (Pass@1)Qwen3 VL 30B A3B Instruct by 4.674.269.6+4.6
- LVBenchQwen3 VL 30B A3B Instruct by 4.562.558+4.5
- MV-BenchQwen3 VL 30B A3B Instruct by 3.672.368.7+3.6
- MMLU-ReduxQwen3 VL 30B A3B Instruct by 3.588.484.9+3.5
- OCRBench v2 (Chinese)Qwen3 VL 8B Instruct by 3.457.861.2+3.4
- VideoMMMUQwen3 VL 30B A3B Instruct by 3.468.765.3+3.4
- ERQAQwen3 VL 8B Instruct by 2.84345.8+2.8
- OCRBench v2_enQwen3 VL 8B Instruct by 2.263.265.4+2.2
- RealWorldQAQwen3 VL 30B A3B Instruct by 2.273.771.5+2.2
- IFEvalQwen3 VL 30B A3B Instruct by 2.185.883.7+2.1
- DocVQA_testQwen3 VL 8B Instruct by 1.19596.1+1.1
- InfoVQA_testQwen3 VL 8B Instruct by 1.18283.1+1.1
- CC-OCRtie80.779.9tie
- AI2Dtie8585.7tie
- AA Intelligencetie7.97.3tie
- HallusionBenchtie61.561.1tie
- MM MTBenchtie8.17.7tie
- ScreenSpottie94.794.4tie
- BFCL v3tie66.366.3tie
Questions people ask
Which is better, Qwen3 VL 30B A3B Instruct or Qwen3 VL 8B Instruct?
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 8B Instruct wins agents and facts, and costs 12% less. They are level on following instructions.
Which is better for coding?
Qwen3 VL 30B A3B Instruct. It wins 3 of the 3 coding tests both models report; Qwen3 VL 8B Instruct wins none.
Which is cheaper?
Qwen3 VL 30B A3B Instruct costs $0.20 per million input tokens and $0.80 per million output tokens; Qwen3 VL 8B Instruct costs $0.18 and $0.70. That makes Qwen3 VL 8B Instruct about 12% cheaper for the same work.
How do you compare the two?
We use the 51 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 32 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.