Developers noticed the effort dial reading low. Anthropic says it is testing serving configs, and the number means something different this week.

Over the weekend of 22 August, Claude Code users posted that the tool's "effort" indicator was showing unexpectedly low values — one screenshot reported 10 out of 100 — and that identical tasks were behaving very differently between Opus 5 and an older Opus 4.6, in one case 43 minutes against 2 minutes for the same job. The thread reached the front page of Hacker News. Thariq, who leads the Claude Code team at Anthropic, replied in public that the company is currently testing API serving configurations and that during that test the numerical effort value is "mapped differently" than usual. He said the effort a user selects is the effort they get, and that the company's own evaluations show no drop in performance. The rest of the thread is not so easily settled: a long run of users describing Opus 5 output as more verbose and more elaborate than they want, some switching back to 4.6 or to rival tools. Those are impressions, not measurements, and impressions of model quality are famously unreliable — but the specific thing Anthropic confirmed is that a number shown in the interface did not mean this week what it meant last week.

✓ Verified · 2 sources

Related stories