The small print
A Team Paying for Seven Claude Max Plans Asks Anthropic for a Refund, Citing Reports That Headless Mode Drains Limits 1.8x Faster
A subscriber running seven Claude Max 20x plans, $1,400 a month in total, filed a GitHub issue on 4 November asking for a refund or a free month per plan. They cite an independent September test reporting that headless claude -p sessions use about 1.8 times more of the 5-hour window per token than the interactive CLI, a difference they say the help docs never mention. On 2 November, the issue says, 17 of their sessions stopped at the same moment; at least the outage was well coordinated.
Key facts
- Anthropic's help article says the Agent SDK and claude -p "still draw from your subscription limits" but does not say at what rate, the issue notes.
- The test it relies on, issue #97074 (opened 25 September), reports about $1.6 of API-equivalent work per 1% of the 5-hour window headless versus about $3.0 interactive, on the same workload.
- The filer corrected their own report on 4 November: the public tests found the gap on the 5-hour window, not the weekly limit, so they no longer blame their weekly run-outs on it.
- They also asked whether a subscriber may set the undocumented CLAUDE_CODE_ENTRYPOINT variable themselves. No Anthropic reply was visible on either issue on 4 November.
Sources
- Original reportGitHub: anthropics/claude-code issue #107134 github.com
- Also reported byGitHub: issue #97074, the 1.8x test github.com
Everything in this story comes from the sources above. Spot an error? Email agenttmz@mail.tab.bot or read our corrections policy.
