Key takeaways

• Qwen3 VL 235B A22B Instruct is an open-weight, multimodal MoE model for unified text generation and visual understanding across images and videos. It targets vision-language tasks such as VQA, document parsing, chart/table extraction, and multilingual OCR.
• The model employs a Mixture of Experts (MoE) architecture with 235 billion parameters and has a context length of 131,072 tokens. It prioritizes robust perception (recognition of diverse real-world and synthetic categories), spatial understanding (2D/3D grounding), and long-form visual comprehension.
• A key strength is its proficiency in general English knowledge and reasoning, achieving 80.71% accuracy on 'AGIEval English'. It also demonstrates strong mathematical problem-solving capabilities, achieving 95.2% accuracy on the 'MATH-500' benchmark.
• The model's major limitations include poor performance in complex reasoning tasks, scoring only 6.402% accuracy on 'Humanity's Last Exam'. It is susceptible to factual inaccuracies, hallucinations, inconsistent answer quality, and failures on questions requiring specific information or negative responses.
• In mathematical contexts, the model struggles with precise interpretation of nuanced geometric diagrams and can make minor arithmetic or logical errors. Conceptual misunderstandings in complex probability scenarios were also noted.

Technical specifications

Context length

131k tokens

Max output

—

Architecture

MoE

Modalities

text+image->text

Parameters

~175B

Open weights

Yes

Licence

Open

Regions

China

Vendor

Qwen

Provider model key

qwen/qwen3-vl-235b-a22b-instruct