AI coding tools · Claude Code ·

Claude 5 hour limit draining fast in T3 Code or Agent SDK

Your Claude 5-hour limit runs out far sooner than usual when you work in T3 Code, an app built on the Claude Agent SDK, or headless claude -p, even though the same kind of work in the Claude Code terminal app lasts much longer. As of October 2026, several users have measured this in public, but Anthropic hasn’t confirmed it or announced a fix. This guide covers what the user reports show and what you can do while they’re open.

If you want the next step, read Claude Code usage limits: how to reduce token usage so they last (Pro and Max).

Quick answer

  • It’s a user-reported metering gap, not a confirmed Anthropic bug. Users measure the same tokens using more of the 5-hour window through the Agent SDK and claude -p than through the interactive Claude Code app.
  • For heavy work, the interactive Claude Code app is the path users measured as cheaper on the 5-hour window.
  • Check your real usage with /usage in Claude Code before you change anything.
  • Add your numbers to the open GitHub issue, so Anthropic sees more than one data point.

What users report

The main report is anthropics/claude-code#97074, opened on 25 Sep 2026 and still open. On a Max plan, its author says the same workload “uses about 1.8× more of the 5-hour window when it runs headless” (claude -p, recorded as sdk-cli) than in the interactive app, with the same token usage. As of 25 Oct 2026, the issue has 20 +1 reactions and 16 comments from 13 accounts. Besides the author, 9 commenters say they saw or measured the same thing, on Pro, Max and Team plans. None of the comments is from Anthropic, and Anthropic hasn’t said whether the difference is a bug or intended.

On 24 Oct 2026, DeepakNess posted on X that after downloading T3 Code, one session on Claude Max “emptied the whole 5-hour limit in ~30 mins”. T3 Code’s maker, Theo, then posted: “I confirmed it’s a real bug on Anthropic’s side. To be clear: this does not affect weekly limits. All Agents SDK usage will drain your 5h limits faster than using Claude Code directly.” That confirmation comes from Theo’s own testing, not from Anthropic. He says he has reported it to Anthropic.

His tracking issue, pingdotgg/t3code#24649, gives his A/B numbers on one Max account: the interactive app used 7% of the 5-hour window per run, T3 Code 13% and 16%, and claude -p 9%, while the weekly meter moved 2% in every run. He puts the gap at about 1.6x to 1.7x for the same tokens, and says T3 Code doesn’t send more tokens or break the prompt cache. These are one person’s tests on one account, not Anthropic’s figures, and the size of the gap varies between reports.

As of 25 Oct 2026, the Claude status page lists no incident about usage limits or the Agent SDK, and we found no statement from Anthropic about it.

Why the Agent SDK uses your plan at all

Anthropic’s help article on using the Claude Agent SDK with your Claude plan says that “Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription limits”. The same article’s 7 October 2026 update says Max and Team plans now “include monthly API credits”, which cover the Claude Agent SDK and claude -p; it doesn’t say how those credits relate to the 5-hour window. It also doesn’t say that this usage is counted at a different rate, which is what the reports ask Anthropic to clarify.

Step 1: check your real usage

Run /usage in Claude Code to see where your 5-hour and weekly limits stand. If you want it on screen all the time, the Claude Code status line docs list rate_limits.five_hour.used_percentage and rate_limits.seven_day.used_percentage among the data passed to your status line script. The docs say rate_limits appears only for claude.ai Pro and Max subscribers, and only after the first response in a session.

Step 2: move heavy work to the interactive app

In every public A/B test we found, the interactive Claude Code app used less of the 5-hour window for the same work than T3 Code, Agent SDK apps or claude -p. Until Anthropic responds, the practical option is to run long or heavy sessions in the interactive app and keep SDK-based tools for lighter jobs. This is a workaround based on user tests, not an official recommendation, and it may stop mattering if Anthropic changes how usage is counted.

Step 3: don’t confuse it with a separate limit report

A different open report, anthropics/claude-code#104738, says an Agent SDK process was rejected with “You’ve hit your session limit” while the Claude desktop app on the same account showed the 5-hour limit at 62%. That’s one user, and it’s a different symptom: two surfaces disagreeing about the limit, rather than one draining faster.

Step 4: add your data to the issue

If you can measure the gap, add your plan, model, Claude Code version and before-and-after 5-hour percentages to #97074. A +1 reaction helps more than a “same here” comment. If your weekly limit is what runs out, see Claude weekly limit reached.

Watch your 5-hour limit in Claude Code, with the kit

Disclosure: this is our own paid product. Usage Limit Saver is a pack of hooks, rules and a weekly routine designed to stretch your Claude Code usage. Its hooks are designed to warn when your 5-hour or weekly plan usage passes a threshold you choose, using Claude Code’s own status-line data, and its usage-log script summarises your 5-hour and weekly percentages over the past days. That data appears only on Pro and Max plans, in interactive sessions, so it’s designed for the interactive app you move heavy work to, not for T3 Code or Agent SDK sessions. It doesn’t raise your limits or change how they’re counted, and it doesn’t fix the gap above. It’s Linux only so far and hasn’t been tested against a live Anthropic account. It costs $12 (plus VAT where applicable).

Sources

As of October 2026 (sources checked 25 Oct 2026):

Claude, Claude Code, Claude Agent SDK, Anthropic, T3 Code, GitHub, X, Linux and other names belong to their owners. Names are used only to identify products and compatibility. Stack & Ship is not affiliated with or endorsed by any of them.

Related posts