Qwen3 VL 30B A3B Instruct vs Qwen3 VL Thinking (8B)
Wins 2 of 7 areas
Coding · Reasoning
Wins 5 of 7 areas
Agents · Facts · Images and charts · Long documents · Following instructions
Qwen3 VL Thinking (8B) is the stronger all-rounder.Qwen3 VL 30B A3B Instruct is cheaper and better at reasoning.
Scores updated · 53 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.
- AgentsCarrying out multi-step tasks on its own02Qwen3 VL Thinking (8B)2 of 2 tests
- FactsGetting facts right instead of making them up02Qwen3 VL Thinking (8B)2 of 2 tests
- Images and chartsUnderstanding pictures, charts and video34Qwen3 VL Thinking (8B)4 of 7 tests
- Long documentsFinding answers in very long texts01Qwen3 VL Thinking (8B)1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3 VL Thinking (8B)1 of 1 test
- 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 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 30B A3B Instruct costs 56% 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 Thinking (8B) pulls ahead
- Recent programming contest problemsLiveCodeBench v6+16points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+6.8points ahead
- Reasons across sets of long documentsAA-LCR+6points ahead
Where Qwen3 VL 30B A3B Instruct pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+11.6points ahead
- Code for real scientific research problemsSciCode+8.9points ahead
- Reads text in imagesOCRBench+8.4points ahead
Every test, side by side
All 53 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3 VL 30B A3B Instruct
- LiveCodeBench v6Qwen3 VL Thinking (8B) by 1642.658.6+16
- SciCodeQwen3 VL 30B A3B Instruct by 8.930.821.9+8.9
- Terminal-Bench HardQwen3 VL 30B A3B Instruct by 2.36.13.8+2.3
AgentsQwen3 VL Thinking (8B)
- GDPValQwen3 VL Thinking (8B) by 8.408.4+8.4
- OSWorld-VerifiedQwen3 VL Thinking (8B) by 3.630.333.9+3.6
ReasoningQwen3 VL 30B A3B Instruct
- GPQA DiamondQwen3 VL 30B A3B Instruct by 11.669.557.9+11.6
- Humanity's Last ExamQwen3 VL 30B A3B Instruct by 2.56.33.8+2.5
- CritPttie00.3tie
FactsQwen3 VL Thinking (8B)
- AA-Omniscience · AccuracyQwen3 VL Thinking (8B) by 4.815.620.4+4.8
- AA-Omniscience · Non-hallucinationQwen3 VL Thinking (8B) by 2.36.38.6+2.3
Images and chartsQwen3 VL Thinking (8B)
- OCRBenchQwen3 VL 30B A3B Instruct by 8.490.381.9+8.4
- MMMU-ProQwen3 VL 30B A3B Instruct by 5.562.156.6+5.5
- CharXiv (RQ)Qwen3 VL Thinking (8B) by 4.148.953+4.1
- MMStarQwen3 VL Thinking (8B) by 3.272.175.3+3.2
- Video-MMEQwen3 VL 30B A3B Instruct by 2.774.571.8+2.7
- MathVistaQwen3 VL Thinking (8B) by 1.380.181.4+1.3
- BLINKQwen3 VL Thinking (8B) by 167.768.7+1
Long documentsQwen3 VL Thinking (8B)
- AA-LCRQwen3 VL Thinking (8B) by 627.333.3+6
Following instructionsQwen3 VL Thinking (8B)
- IFBenchQwen3 VL Thinking (8B) by 6.833.139.9+6.8
Other results34 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.
- SimpleQAQwen3 VL Thinking (8B) by 22.62749.6+22.6
- ScreenSpot-Pro (No tools)Qwen3 VL 30B A3B Instruct by 13.960.546.6+13.9
- AA-OmniscienceQwen3 VL Thinking (8B) by 11.1-63.4-52.3+11.1
- AIME 2025Qwen3 VL Thinking (8B) by 1169.380.3+11
- HMMT 2025Qwen3 VL Thinking (8B) by 1050.660.6+10
- Arena-Hard-V2Qwen3 VL 30B A3B Instruct by 7.458.551.1+7.4
- LVBenchQwen3 VL 30B A3B Instruct by 6.762.555.8+6.7
- AA Agentic IndexQwen3 VL Thinking (8B) by 6.19.515.6+6.1
- Artificial Analysis Coding IndexQwen3 VL 30B A3B Instruct by 4.514.39.8+4.5
- CC-OCRQwen3 VL 30B A3B Instruct by 4.480.776.3+4.4
- VideoMMMUQwen3 VL Thinking (8B) by 4.168.772.8+4.1
- InfoVQA_testQwen3 VL Thinking (8B) by 48286+4
- HallusionBenchQwen3 VL Thinking (8B) by 3.961.565.4+3.9
- ERQAQwen3 VL Thinking (8B) by 3.84346.8+3.8
- CharadesSTAQwen3 VL 30B A3B Instruct by 3.663.559.9+3.6
- τ²-Bench Telecom (AA run)Qwen3 VL Thinking (8B) by 3.51922.5+3.5
- BFCL v3Qwen3 VL 30B A3B Instruct by 3.366.363+3.3
- MV-BenchQwen3 VL 30B A3B Instruct by 3.372.369+3.3
- IFEvalQwen3 VL 30B A3B Instruct by 2.685.883.2+2.6
- MATH-VisionQwen3 VL Thinking (8B) by 2.560.262.7+2.5
- IncludeQwen3 VL 30B A3B Instruct by 2.171.669.5+2.1
- SuperGPQAQwen3 VL 30B A3B Instruct by 1.953.151.2+1.9
- OCRBench v2 (Chinese)Qwen3 VL Thinking (8B) by 1.457.859.2+1.4
- ScreenSpotQwen3 VL 30B A3B Instruct by 1.194.793.6+1.1
- OCRBench v2_entie63.263.9tie
- MMLU-Protie77.877.3tie
- MMLU-Reduxtie88.488.8tie
- AA Intelligencetie7.98.2tie
- DocVQA_testtie9595.3tie
- MMLU-ProXtie70.970.7tie
- RealWorldQAtie73.773.5tie
- AI2Dtie8584.9tie
- MM MTBenchtie8.18tie
- MMMU (val) (Pass@1)tie74.274.1tie
Questions people ask
Which is better, Qwen3 VL 30B A3B Instruct or Qwen3 VL Thinking (8B)?
Qwen3 VL Thinking (8B) wins five of the seven areas where both have results: agents, facts, images and charts, long documents and following instructions. Qwen3 VL 30B A3B Instruct wins coding and reasoning, and costs 56% less.
Which is better for coding?
Qwen3 VL 30B A3B Instruct. It wins 2 of the 3 coding tests both models report; Qwen3 VL Thinking (8B) wins 1.
Which is cheaper?
Qwen3 VL 30B A3B Instruct costs $0.20 per million input tokens and $0.80 per million output tokens; Qwen3 VL Thinking (8B) costs $0.18 and $2.10. That makes Qwen3 VL 30B A3B Instruct about 56% cheaper for the same work.
How do you compare the two?
We use the 53 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 34 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.