GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2

Recent AI benchmarks reveal that larger models don't necessarily perform better in real-world accuracy, with MIT-licensed GLM-5.2 (753B parameters) scoring nearly as high as proprietary GPT-5.5 (estimated 1-2T parameters) while maintaining a 28% hallucination rate compared to GPT-5.5's 86%. The findings suggest that massive parameter scaling has hit diminishing returns, and larger models often fail to learn when to admit uncertainty, instead confidently generating incorrect answers on complex technical questions.