Claude Opus 4.8

Claude Opus 4.8 quality drops 8.6 points with coding decline

Claude Opus 4.8 shows a significant performance decline in the current benchmark window, with overall quality dropping from 97.7 to 89.1 points. The most notable regression appears in coding performance, which fell from 94 to 88 points. Reasoning capability data is conspicuously absent from the current window despite scoring a perfect 100 previously, while creative writing scores at 80 represent a new category without historical comparison. Multilingual performance remains the model's strongest suit, holding steady at 99 points across both windows. Latency improved marginally from 7820ms to 7692ms at the median, showing slightly faster response times. The quality drop of 8.6 points is substantial enough to warrant attention from users who depend on consistent performance, particularly those relying on coding assistance. The missing reasoning scores and appearance of new creative scores suggest possible changes to the benchmark methodology or model capabilities between windows. Users should monitor whether this represents temporary instability or a sustained regression in model quality.