Z.ai: GLM 4.7

Z.AI's GLM-4.7, released December 21, 2025, is positioned as a premium model with a context length of 200,000. It exhibits exceptional reliability, boasting a 98% success rate across benchmarks, indicating consistent operational stability. However, its speed is a notable weakness, with response times ranking in the 5th percentile. Similarly, its pricing is at the premium end, ranking in the 12th percentile. Despite its slower speed and higher cost, GLM-4.7 demonstrates strong performance in several key areas. It excels in Instruction Following (91st percentile accuracy), Reasoning (80th percentile accuracy), and Mathematics (82nd percentile accuracy), aligning with its description of enhanced programming capabilities and stable multi-step reasoning. Its perfect 100% accuracy in Ethics is particularly impressive, making it the most accurate model at its price point and among models of comparable speed. General Knowledge is also solid at 99.5% accuracy (70th percentile). A key area for improvement is its hallucination rate, with 88.0% accuracy placing it in the 36th percentile, suggesting it could better acknowledge uncertainty.