Opus 5 Shows Substantial Gains in Mathematics and Visual Understanding

Comparative testing between Opus 5 and its predecessor Opus 4.8 demonstrates substantial improvements in three major categories. Mathematical research on Riemann-bench jumped from 47.2% to 68%, a gain of 21 percentage points. Chart understanding also improved significantly on Chartography, rising from 15.9% to 27.3%. Long-context agents demonstrated similar progress, increasing from 21.9% to 32.3% on HANDBOOK.md. Other categories including creative writing, document reasoning, and everyday chat showed modest gains or equivalent performance.