Qwen: Qwen3 Coder 30B A3B Instruct

Qwen3 Coder 30B A3B Instruct demonstrates a strong overall performance profile, particularly excelling in its intended domain of code-related tasks. The model performs among the fastest models, ranking in the 68th percentile for speed, and offers competitive pricing, placing in the 76th percentile for cost-effectiveness. Its reliability is notably high, with a 92% success rate across benchmarks, indicating consistent and usable responses. In terms of specific benchmarks, Qwen3 Coder 30B A3B Instruct achieved perfect accuracy in Keyword Topic Relevance Classification, showcasing excellent semantic understanding and making it the most accurate model at its price point and among models of comparable speed. It also performed very strongly in Coding (89.0% accuracy) and Mathematics (89.0% accuracy), aligning with its design for advanced code generation. The model exhibits good ethical reasoning (98.0% accuracy) and a solid ability to acknowledge uncertainty, with a 94.0% accuracy in hallucination detection. However, its performance in General Knowledge (32.2% accuracy) and Email Classification (91.0% accuracy, 16th percentile) is comparatively weaker. Instruction Following also presents an area for improvement at 51.5% accuracy. Overall, its strengths lie in specialized technical and classification tasks, while general knowledge and nuanced instruction following could be further refined.