Qwen: Qwen3 VL 235B A22B Thinking

Qwen3 VL 235B A22B Thinking consistently ranks among the fastest models and offers highly competitive pricing across various benchmarks. This multimodal model, designed for strong text generation and visual understanding, demonstrates notable strengths in specific areas. It achieves a high accuracy of 95.2% in Hallucinations (Baseline) testing, indicating a strong ability to acknowledge uncertainty, and an impressive 81.6% in Instruction Following (Baseline), showcasing its precision in adhering to complex directives. The model also performs well in Email Classification (Baseline) with 97.8% accuracy. However, the model exhibits significant weaknesses in core cognitive benchmarks. It scores 0.0% accuracy in Coding (Baseline), General Knowledge (Baseline), and Ethics (Baseline), suggesting a lack of foundational understanding or an inability to perform well in these specific multiple-choice formats. Its performance in Reasoning (Baseline) and Mathematics (Baseline) is also very low, at 3.4% and 3.1% accuracy respectively, despite its description emphasizing optimization for multimodal reasoning in STEM and math. While its speed and cost efficiency are exceptional, the model's current performance across several critical reasoning and knowledge-based tasks indicates areas requiring substantial improvement.