Qwen: Qwen3 30B A3B Instruct 2507

Qwen3 30B A3B Instruct 2507, a 30.5B-parameter mixture-of-experts model with 3.3B active parameters, demonstrates strong overall performance, particularly in speed and reliability. It performs among the fastest models, typically ranking in the top tier (65th percentile), and offers competitive pricing (59th percentile). Notably, the model exhibits exceptional reliability with a 100% success rate across all benchmarks, indicating minimal technical failures. The model excels in specific areas, achieving perfect accuracy in Keyword Topic Relevance Classification and Ethics (Baseline), where it is also identified as the most accurate model at its price point and speed. It shows strong performance in Mathematics (94.0% accuracy, 83rd percentile) and Coding (93.0% accuracy, 82nd percentile), aligning with its design for high-quality instruction following and agentic tool use. Its ability to appropriately acknowledge uncertainty is high, with 98.0% accuracy in Hallucinations (Baseline). While General Knowledge (97.0% accuracy) and Reasoning (78.0% accuracy) are solid, Email Classification (94.0% accuracy, 25th percentile) and Instruction Following (59.8% accuracy, 56th percentile) represent areas for potential improvement compared to its other strengths. Overall, Qwen3 30B A3B Instruct 2507 is a robust model with competitive speed, price, and outstanding reliability, making it a strong contender for tasks requiring high-quality instruction following and specialized reasoning.