Claude Code usage draining fast? Anthropic found no regression, blames heavy sessions and ships smaller fixes

Anthropic says it has found no single bug behind complaints that Claude Code was eating through usage limits faster than usual. Boris Cherny, who works on Claude Code at Anthropic, posted the results of the team's review on October 28 at 23:51 UTC, a week after the complaints started.

The thread began on October 21, when a user wrote that usage was "draining 5x faster" than on any model before. The next day Cherny replied that there had been "no changes on our end" and asked for the output of claude -p /usage; users then shared session IDs.

Having gone through those sessions, he now reports three things:

• the team found "a few smaller things" that made some sessions burn tokens faster than they should have, and fixes for those are rolling out now;
• there was no single root cause and no regression in Claude Code itself;
• in most cases, users who hit limits faster could trace it to one or two heavy sessions using xhigh effort, /goal, Workflows, Agent Teams or long SDK jobs.

The most useful detail is a tip near the end: subagents inherit the parent session's effort level by default, "and it can add up fast." A session running at xhigh that fans out into several subagents runs all of them at xhigh too. Cherny points to /usage to see where a session's tokens went, and to /feedback for anything that still looks wrong.

The post doesn't list the fixes or the Claude Code version that carries them, so there is no way yet to tell how much of the reported drain they will recover. Cherny framed the rest as a communication problem rather than a code one: "it's on us to make it clearer when to use them and how much these features cost." He says the team will follow up directly with everyone who shared a session ID.