Qwen: Qwen3 4B

Qwen3 4B, a 4 billion parameter model from qwen, demonstrates moderate speed performance, ranking in the 20th percentile across benchmarks. It offers competitive pricing, positioned at the 50th percentile. Notably, the model exhibits exceptional reliability with a 98% success rate, indicating minimal technical failures and consistent response generation. In terms of performance across categories, Qwen3 4B shows strong capabilities in Reasoning (96.0% accuracy, 87th percentile) and General Knowledge (99.0% accuracy, 66th percentile), suggesting proficiency in complex problem-solving and broad factual recall. Its Mathematics performance is also solid at 89.0% accuracy (58th percentile). However, the model struggles with Instruction Following (44.9% accuracy, 39th percentile) and Hallucinations (82.0% accuracy, 28th percentile), indicating areas for improvement in adhering to complex directives and acknowledging uncertainty. Email Classification also presents a weakness with 93.0% accuracy (22nd percentile). Its dual-mode architecture aims to balance high-precision reasoning with efficient dialogue, making it suitable for multi-turn chat and agent workflows despite some accuracy limitations in specific tasks.