Opus 4.6/4.7 fills the mathematical shortcoming to match coding ability. The ECI metric is interesting—it measures the relative difficulty between models, not compared to humans, so don't mistake it for "surpassing human mathematicians."

Epoch AI's domain-specific capability index analysis shows that the Claude series has long been a leader in coding, but has relatively weaker math skills; recent data indicates this imbalance is rapidly improving. Previously, Claude's SWE-ECI was higher than the overall score, and there was a gap in Math-ECI, but Opus 4.6/4.7 narrowed both to within 1 point, addressing the shortcomings. ECI reflects the relative task difficulty between models, not the difficulty for humans.