Z.ai: GLM 4 32B

Z.ai's GLM 4 32B model, created on July 24, 2025, demonstrates competitive response times, ranking in the 53rd percentile for speed across eight benchmarks. It consistently offers highly competitive pricing, placing in the 84th percentile. The model excels in specific areas, particularly in its ability to acknowledge uncertainty, achieving 98.0% accuracy in Hallucinations (Baseline) tests, indicating a strong capacity to avoid generating fabricated information. It also performs well in Email Classification, with 98.0% accuracy. However, GLM 4 32B exhibits significant weaknesses in complex cognitive tasks. Its performance in Ethics (9.0% accuracy), Mathematics (30.0% accuracy), Reasoning (12.0% accuracy), and Coding (2.0% accuracy) is notably low, placing it in the lower percentiles for these categories. General Knowledge also presents a challenge, with 90.5% accuracy but a 27th percentile ranking. While it shows moderate capability in Instruction Following (55.0% accuracy), its overall strength lies in cost-effectiveness and its ability to handle straightforward classification and uncertainty acknowledgment, rather than advanced problem-solving or ethical reasoning.