Qwen: Qwen3 235B A22B Thinking 2507

Qwen3-235B-A22B-Thinking-2507 is a high-performance, open-weight Mixture-of-Experts (MoE) model optimized for complex reasoning. While demonstrating exceptional reliability with a 100% success rate across all benchmarks, its speed performance tends to be slower, ranking in the 14th percentile. Similarly, its pricing is positioned at premium levels, ranking in the 4th percentile. The model exhibits significant strengths in specialized areas. It achieves perfect accuracy in General Knowledge, making it the most accurate model at its price point and among models of comparable speed. Its Coding performance is outstanding at 98.0% accuracy (99th percentile), and it excels in Reasoning (98.0% accuracy, 87th percentile) and Mathematics (92.9% accuracy, 68th percentile), aligning with its "thinking-only" optimization. Hallucination rates are low at 94.0% accuracy. However, a notable weakness is its Instruction Following capability, which is relatively low at 26.3% accuracy (23rd percentile). Email Classification and Keyword Topic Relevance Classification show solid, though not top-tier, performance at 99.0% and 90.0% accuracy respectively. Ethics performance is moderate at 98.0% accuracy (34th percentile).