Comparing AI model performance requires accounting for data variance

Original: You have to look at Sonnet 5 in the context of GLM, not just Opus models.

Source: x.com ↗