Theo says Anthropic's Agents SDK burns 5-hour limits faster
One user emptied a full 5-hour window in about 30 minutes, and Theo thinks a recent limit boost never reached the SDK.
Theo says he confirmed a real bug on Anthropic's side. All Agents SDK usage drains your 5-hour limits faster than using Claude Code directly, he wrote, and weekly limits are not affected. He says he has reported it to Anthropic and is hoping for a fix soon.
All Agents SDK usage will drain your 5h limits faster than using Claude Code directly.
His guess at the cause is mundane. He suspects it comes from the recent boost to 5-hour limits, and that whoever applied it forgot to apply it to the Agents SDK. He points to a GitHub issue opened a few days after that change as supporting evidence, and puts himself at roughly 95% confident.
The report that started it
The thread Theo was responding to came from @deepaknesss, who said he downloaded T3 Code and hit the 5-hour limit on his Claude Max plan for the first time. It was a single session asking for fixes to a few payment related bugs, and the window was gone in about half an hour. He said he prompted exactly the way he prompts inside the Claude Code CLI, and asked whether T3 Code works differently or whether he had made a mistake.
That is the shape of the whole thing. One user noticed his quota evaporating on a workload that would not have done that in Claude Code, and the explanation turned out to be upstream of any particular client.
What this means if you build on the SDK
The Agents SDK is what you use when you want Claude's agent loop inside your own tool rather than in Anthropic's CLI. Plenty of the harnesses people are running right now sit on top of it, which is why a single accounting discrepancy shows up as "this new tool is burning my plan" rather than "Anthropic's limit boost has a gap in it."
So if you are on a Max plan, running something built on the SDK, and hitting the 5-hour wall sooner than your usage feels like it should, the arithmetic may be wrong rather than your workload. Weekly limits, per Theo, are fine.
Two caveats worth holding. The confirmation here is Theo's own, based on a day of testing plus a GitHub issue he reads as corroboration, and he is the one saying he is 95% sure. There is no statement from Anthropic in the sources, no fix, and no timeline. Until one lands, the practical move is to watch which harness you were in when the window emptied.
it emptied the whole 5-hour limit in ~30 mins
Have the good opinion, not the loud one
One email a day. Under three minutes. What working engineers found out about coding with AI.

