THUDM: GLM 4 32B

THUDM: GLM 4 32B, a 32B bilingual model optimized for code generation, function calling, and agent tasks, consistently performs among the fastest models and offers highly competitive pricing across various benchmarks. Created on April 17, 2025, with a 32000 context length, it demonstrates strong performance in specific areas. The model excels in Email Classification, achieving 99.0% accuracy (87th percentile), and shows good capability in handling Hallucinations with 96.0% accuracy (58th percentile). However, it exhibits significant weaknesses in complex reasoning tasks, scoring only 6.1% in Reasoning (7th percentile), and struggles severely with Instruction Following (0.0% accuracy) and Coding (1.0% accuracy). Its performance in General Knowledge is moderate at 93.5% (38th percentile), while its Ethics score is notably low at 11.0% (10th percentile). Overall, GLM 4 32B presents a mixed profile, showcasing strengths in specific classification and hallucination avoidance, but requiring substantial improvement in areas demanding intricate logical processing, ethical understanding, and precise instruction adherence.