Opus 4.8 dropped a few hours ago. The announcement is, predictably, all benchmark deltas and SWE-bench numbers. The decision teams actually have to make this week is not 'is 4.8 better than 4.7' — it is 'which of my running workloads should move, which should stay, and what is the regression risk on the ones I move'. Here is the upgrade decision tree I am giving my own team, with three workload types I am explicitly keeping on 4.7 until at least mid-July.

Anthropic shipped Opus 4.8 this week — three Opus versions in less than four months. Most coverage will benchmark the model. The buried story is what the release cadence does to…

Anthropic's new Claude Opus 4.8 aced our math problem and shipped a spotless game—then drained our entire token quota in a single prompt.