Anthropic released Claude Opus 4.8 on 28 May. The company highlighted improvements in coding, agentic tasks, reasoning, and professional work, while keeping the listed model price unchanged. Claude users also gained more control over the effort used for a task.
Quick exploration and easy transformations.
Difficult analysis with a clear quality test.
A human decision before the next costly stage.
Long-running work needs visible checkpoints
More effort can help on a difficult problem, but it can also add time and cost without improving an easy one. Long-running work introduces another risk: a wrong assumption made early may remain hidden until the final result.
What to test for yourself
Split a large task into reviewable stages and define what evidence is required at each one. Compare two effort levels on the same work, then measure correctness, elapsed time, token cost, and human repair rather than judging the longer answer as automatically better.
Create stages with a visible deliverable.
Define sources and tests for each stage.
Run the same task at two suitable settings.
Include time, cost, retries, and corrections.
The useful promise of Opus 4.8 was not that it could run longer. It was that it might stay reliable while doing so. Teams still needed repeated tests to know whether that consistency appeared in their own workflow.
