Qwen3 VL 32B Instruct vs Qwen3 VL Thinking (8B)
Wins 4 of 7 areas
Coding · Reasoning · Images and charts · Long documents
Wins 2 of 7 areas
Agents · Facts
Qwen3 VL 32B Instruct wins more areas, narrowly.Qwen3 VL Thinking (8B) is better at agents.
Scores updated · 49 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 32B Instruct5 of 6 tests
- ReasoningHard problems that need careful thinking20Qwen3 VL 32B Instruct2 of 3 tests · 1 tie
- CodingWriting and fixing software21Qwen3 VL 32B Instruct2 of 3 tests
- Long documentsFinding answers in very long texts10Qwen3 VL 32B Instruct1 of 1 test
- AgentsCarrying out multi-step tasks on its own02Qwen3 VL Thinking (8B)2 of 2 tests
- FactsGetting facts right instead of making them up01Qwen3 VL Thinking (8B)1 of 2 tests · 1 tie
- 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 32B Instruct costs 65% 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 32B Instruct pulls ahead
- Reasoning about charts from research papersCharXiv (RQ)+9.8points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+9.2points ahead
- Code for real scientific research problemsSciCode+8.2points ahead
Where Qwen3 VL Thinking (8B) pulls ahead
- Recent programming contest problemsLiveCodeBench v6+14.8points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+5.6points ahead
- Visual perception tasks people solve at a glanceBLINK+1.4points ahead
Every test, side by side
All 49 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3 VL 32B Instruct
- LiveCodeBench v6Qwen3 VL Thinking (8B) by 14.843.858.6+14.8
- SciCodeQwen3 VL 32B Instruct by 8.230.121.9+8.2
- Terminal-Bench HardQwen3 VL 32B Instruct by 4.58.33.8+4.5
AgentsQwen3 VL Thinking (8B)
- GDPValQwen3 VL Thinking (8B) by 8.408.4+8.4
- OSWorld-VerifiedQwen3 VL Thinking (8B) by 1.332.633.9+1.3
ReasoningQwen3 VL 32B Instruct
- GPQA DiamondQwen3 VL 32B Instruct by 9.267.157.9+9.2
- Humanity's Last ExamQwen3 VL 32B Instruct by 36.83.8+3
- CritPttie00.3tie
FactsQwen3 VL Thinking (8B)
- AA-Omniscience · AccuracyQwen3 VL Thinking (8B) by 5.614.820.4+5.6
- AA-Omniscience · Non-hallucinationtie8.28.6tie
Images and chartsQwen3 VL 32B Instruct
Long documentsQwen3 VL 32B Instruct
- AA-LCRQwen3 VL 32B Instruct by 134.333.3+1
Other results31 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 Thinking (8B) by 14.166.280.3+14.1
- Arena-Hard-V2Qwen3 VL 32B Instruct by 13.664.751.1+13.6
- ScreenSpot-Pro (No tools)Qwen3 VL 32B Instruct by 11.357.946.6+11.3
- AA-OmniscienceQwen3 VL Thinking (8B) by 11.1-63.4-52.3+11.1
- LVBenchQwen3 VL 32B Instruct by 863.855.8+8
- BFCL v3Qwen3 VL 32B Instruct by 7.270.263+7.2
- τ²-Bench Telecom (AA run)Qwen3 VL 32B Instruct by 6.729.222.5+6.7
- AA Agentic IndexQwen3 VL Thinking (8B) by 5.99.715.6+5.9
- Artificial Analysis Coding IndexQwen3 VL 32B Instruct by 5.815.69.8+5.8
- RealWorldQAQwen3 VL 32B Instruct by 5.57973.5+5.5
- AI2DQwen3 VL 32B Instruct by 4.689.584.9+4.6
- IncludeQwen3 VL 32B Instruct by 4.57469.5+4.5
- CC-OCRQwen3 VL 32B Instruct by 480.376.3+4
- MV-BenchQwen3 VL 32B Instruct by 3.872.869+3.8
- OCRBench v2_enQwen3 VL 32B Instruct by 3.567.463.9+3.5
- SuperGPQAQwen3 VL 32B Instruct by 3.454.651.2+3.4
- MMLU-ProXQwen3 VL 32B Instruct by 2.773.470.7+2.7
- ScreenSpotQwen3 VL 32B Instruct by 2.295.893.6+2.2
- ERQAQwen3 VL 32B Instruct by 248.846.8+2
- MMMU (val) (Pass@1)Qwen3 VL 32B Instruct by 1.97674.1+1.9
- DocVQA_testQwen3 VL 32B Instruct by 1.696.995.3+1.6
- HallusionBenchQwen3 VL Thinking (8B) by 1.663.865.4+1.6
- IFEvalQwen3 VL 32B Instruct by 1.584.783.2+1.5
- CharadesSTAQwen3 VL 32B Instruct by 1.361.259.9+1.3
- MMLU-ProQwen3 VL 32B Instruct by 1.378.677.3+1.3
- InfoVQA_testQwen3 VL 32B Instruct by 18786+1
- MMLU-ReduxQwen3 VL 32B Instruct by 189.888.8+1
- MATH-Visiontie63.462.7tie
- MM MTBenchtie8.48tie
- AA Intelligencetie8.48.2tie
- OCRBench v2 (Chinese)tie59.259.2tie
Questions people ask
Which is better, Qwen3 VL 32B Instruct or Qwen3 VL Thinking (8B)?
Qwen3 VL 32B Instruct wins four of the seven areas where both have results: coding, reasoning, images and charts and long documents. Qwen3 VL Thinking (8B) wins agents and facts. They are level on following instructions.
Which is better for coding?
Qwen3 VL 32B Instruct. It wins 2 of the 3 coding tests both models report; Qwen3 VL Thinking (8B) wins 1.
Which is cheaper?
Qwen3 VL 32B Instruct costs $0.16 per million input tokens and $0.64 per million output tokens; Qwen3 VL Thinking (8B) costs $0.18 and $2.10. That makes Qwen3 VL 32B Instruct about 65% cheaper for the same work.
How do you compare the two?
We use the 49 benchmark tests both models have published scores on. The verdict counts the 18 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 31 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.