Qwen3 VL 235B A22B Instruct vs Step3 VL 10B
Wins 4 of 7 areas
Coding · Agents · Images and charts · Long documents
Wins 1 of 7 areas
Following instructions
Qwen3 VL 235B A22B Instruct wins more areas, narrowly.Step3 VL 10B is better at following instructions.
Scores updated · 25 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 video30Qwen3 VL 235B A22B Instruct3 of 5 tests · 2 ties
- CodingWriting and fixing software20Qwen3 VL 235B A22B Instruct2 of 2 tests
- AgentsCarrying out multi-step tasks on its own10Qwen3 VL 235B A22B Instruct1 of 1 test
- Long documentsFinding answers in very long texts10Qwen3 VL 235B A22B Instruct1 of 1 test
- Following instructionsDoing exactly what it is asked01Step3 VL 10B1 of 1 test
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
- FactsGetting facts right instead of making them up11Even1 each
Agents, 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 235B A22B Instruct pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+7points ahead
- Code for real scientific research problemsSciCode+4.8points ahead
- Visual perception tasks people solve at a glanceBLINK+3.9points ahead
Where Step3 VL 10B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+8.3points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+7.5points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+4.2points ahead
Every test, side by side
All 25 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3 VL 235B A22B Instruct
- SciCodeQwen3 VL 235B A22B Instruct by 4.835.931.1+4.8
- Terminal-Bench HardQwen3 VL 235B A22B Instruct by 1.56.85.3+1.5
AgentsQwen3 VL 235B A22B Instruct
- GDPValQwen3 VL 235B A22B Instruct by 6.76.70+6.7
ReasoningEven
- Humanity's Last ExamStep3 VL 10B by 4.26.610.8+4.2
- GPQA DiamondQwen3 VL 235B A22B Instruct by 2.271.269+2.2
- CritPttie00tie
FactsEven
- AA-Omniscience · Non-hallucinationStep3 VL 10B by 8.38.416.7+8.3
- AA-Omniscience · AccuracyQwen3 VL 235B A22B Instruct by 720.313.3+7
Images and chartsQwen3 VL 235B A22B Instruct
Long documentsQwen3 VL 235B A22B Instruct
- AA-LCRQwen3 VL 235B A22B Instruct by 32.732.70+32.7
Following instructionsStep3 VL 10B
- IFBenchStep3 VL 10B by 7.542.750.2+7.5
Other results10 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.
- HMMT 2025Step3 VL 10B by 34.757.492.1+34.7
- τ²-Bench Telecom (AA run)Qwen3 VL 235B A22B Instruct by 1935.116.1+19
- AA Agentic IndexQwen3 VL 235B A22B Instruct by 13.719.15.4+13.7
- AIME 2025Step3 VL 10B by 1374.787.7+13
- ScreenSpot-Pro (No tools)Qwen3 VL 235B A22B Instruct by 10.46251.5+10.4
- AA-OmniscienceQwen3 VL 235B A22B Instruct by 6.3-52.7-59+6.3
- MATH-VisionStep3 VL 10B by 4.366.570.8+4.3
- Artificial Analysis Coding IndexQwen3 VL 235B A22B Instruct by 2.616.513.9+2.6
- AA IntelligenceQwen3 VL 235B A22B Instruct by 2.29.97.7+2.2
- AI2Dtie89.789.3tie
Questions people ask
Which is better, Qwen3 VL 235B A22B Instruct or Step3 VL 10B?
Qwen3 VL 235B A22B Instruct wins four of the seven areas where both have results: coding, agents, images and charts and long documents. Step3 VL 10B wins following instructions. They are level on reasoning and facts.
Which is better for coding?
Qwen3 VL 235B A22B Instruct. It wins 2 of the 2 coding tests both models report; Step3 VL 10B wins none.
How do you compare the two?
We use the 25 benchmark tests both models have published scores on. The verdict counts the 15 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 10 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.