THUDM: GLM 4 32B

THUDM: GLM 4 32B, created on April 17, 2025, is a 32B bilingual model optimized for code generation, function calling, and agent-style tasks. It consistently performs among the fastest models and offers highly competitive pricing across all evaluated benchmarks. In terms of specific performance, the model demonstrates exceptional capability in Email Classification, achieving 99.0% accuracy, placing it in the 87th percentile. It also performs strongly in Reasoning tasks with 74.0% accuracy (74th percentile), indicating robust complex problem-solving abilities. General Knowledge is another area of strength, with 93.5% accuracy, though its percentile ranking is moderate. However, the model exhibits significant weaknesses in other critical areas. Its performance in Coding (Baseline) is notably poor, with only 1.0% accuracy (8th percentile), suggesting it is not suitable for this task despite its description. Similarly, Instruction Following yielded 0.0% accuracy, indicating a complete failure in this domain. Ethics (Baseline) also shows very low accuracy at 11.0% (11th percentile). While its speed is a consistent advantage, the model's high duration in Coding and Ethics benchmarks suggests inefficiency in these specific, challenging tasks. Overall, GLM 4 32B excels in classification and reasoning but struggles considerably with coding, instruction following, and ethical considerations.