Mage-VL-4B is the stronger all-rounder.
Scores updated · 27 tests both models report · How we compare
Where each one wins
Tests won in each of the one area 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.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Mage-VL-4B pulls ahead
Where Qwen3 VL 4B (Reasoning) pulls ahead
- Questions that truly need the image to answerMMStar+5.9points ahead
Every test, side by side
All 27 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Images and chartsMage-VL-4B
Other results23 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.
- Ref-DAVIS17Mage-VL-4B by 18.325.87.5+18.3
- Timelens-ActivityNetMage-VL-4B by 1745.428.4+17
- LVBenchQwen3 VL 4B (Reasoning) by 11.741.853.5+11.7
- SATMage-VL-4B by 10.667.356.7+10.6
- VSI-BenchMage-VL-4B by 9.164.355.2+9.1
- MLVU-devMage-VL-4B by 7.268.761.5+7.2
- RT-AvgMage-VL-4B by 779.872.8+7
- BT-AvgQwen3 VL 4B (Reasoning) by 4.948.153.1+4.9
- MV-BenchQwen3 VL 4B (Reasoning) by 4.265.169.3+4.2
- ChartQAProQwen3 VL 4B (Reasoning) by 3.632.636.2+3.6
- LongVideoBenchMage-VL-4B by 3.661.357.7+3.6
- MME-RealWorldMage-VL-4B by 3.366.563.2+3.3
- NextQAMage-VL-4B by 3.383.179.8+3.3
- TextVQA-valQwen3 VL 4B (Reasoning) by 3.377.380.5+3.3
- CV-BenchMage-VL-4B by 2.187.885.7+2.1
- AI2D w/ MaskMage-VL-4B by 1.783.281.5+1.7
- ChartQAMage-VL-4B by 1.684.983.3+1.6
- MMBench-CN-devMage-VL-4B by 1.48280.6+1.4
- InfoVQA (val)tie80.379.5tie
- MMBench-EN-devtie8483.3tie
- MME-Perceptiontie17101704tie
- DocVQA-valtie95.194.7tie
- MultiDocVQA-valtie87.587.2tie
Questions people ask
Which is better, Mage-VL-4B or Qwen3 VL 4B (Reasoning)?
Mage-VL-4B wins the one area where both have results: images and charts. Qwen3 VL 4B (Reasoning) wins none.
How do you compare the two?
We use the 27 benchmark tests both models have published scores on. The verdict counts the 4 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 23 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.