Baidu: Qianfan-OCR-Fast

Baidu's Qianfan-OCR-Fast, a domain-specific multimodal large model for OCR, demonstrates competitive response times, performing among the average models with a 56th percentile speed ranking. Its pricing is also competitive, ranking in the 54th percentile. A significant strength of this model is its exceptional reliability, boasting a 97% success rate across benchmarks, indicating minimal technical failures. In terms of performance across categories, Qianfan-OCR-Fast shows notable strengths in Ethics and Email Classification, achieving 98.0% and 96.6% accuracy respectively, though these still place it in the 43rd and 35th percentiles. Its Hallucinations accuracy is 76.2%, which is relatively low at the 20th percentile, suggesting a tendency to provide answers even when uncertain. The model struggles with Reasoning (35.7% accuracy, 18th percentile) and Instruction Following (44.1% accuracy, 33rd percentile), indicating areas for improvement in complex logical and multi-step task execution. General Knowledge, Coding, and Mathematics performance are moderate, hovering around the 33rd percentile. Overall, while highly reliable and competitively priced, its accuracy across many core benchmarks suggests it may not be a top performer in general AI tasks, but its specialized OCR focus implies its true strength lies in that domain.