benchgap
Alibaba (Qwen) · model

Qwen3-VL-235B-A22B Instruct benchmark scores

As of 2026-11-19, Qwen3-VL-235B-A22B Instruct (Alibaba (Qwen)) has measured scores on 8 benchmarks and estimated scores on 20 more.

BenchmarkScoreSource
AA-MMMU-Pro67.4%estimated ± 1.5 pp, high confidence
MMMU78.7%measured
ParseBench-150.0%estimated ± 5.2 pp, low confidence
ParseBench-255.7%estimated ± 5.0 pp, medium confidence
CharXiv72.5%estimated ± 3.3 pp, high confidence
CharXiv w/o tools74.2%estimated ± 2.1 pp, high confidence
AI2D_TEST87.8%estimated ± 1.4 pp, medium confidence
CountBench97.4%estimated ± 1.6 pp, low confidence
DynaMath79.1%estimated ± 1.0 pp, medium confidence
ERQA53.8%estimated ± 4.2 pp, high confidence
MathVision76.0%estimated ± 2.9 pp, high confidence
MLVU (M-Avg)81.5%estimated ± 2.2 pp, high confidence
MMMU-Pro68.1%measured
OmniDocBench 1.590.7%estimated ± 1.5 pp, low confidence
RealWorldQA79.3%measured
ScreenSpot Pro62.0%measured
SimpleVQA59.2%estimated ± 7.1 pp, medium confidence
V*90.9%estimated ± 6.5 pp, low confidence
Video-MME (w/o subtitle)79.2%measured
Video-MME (with subtitle)82.0%estimated ± 3.0 pp, high confidence
VideoMMMU74.7%measured
We-Math75.6%estimated ± 5.5 pp, low confidence
MedXpertQA (MM)46.1%estimated ± 2.4 pp, medium confidence
MMVU70.9%estimated ± 2.7 pp, medium confidence
CC-OCR82.2%measured
RefCOCO (avg)93.5%estimated ± 1.8 pp, low confidence
OCRBench V267.1%measured
MMMU-Pro w/ Python74.1%estimated ± 2.6 pp, high confidence