What changed
Before Claude Code v2.1.234, hitting a usage limit mid-session was a hard stop. You'd see a message naming the reset time, and then nothing happened on its own, you had to notice the window had reopened and manually pick the session back up. Released 17 August 2026, v2.1.234 changed that default. Per Claude Code's own changelog:
Claude Code now continues your session automatically when a claude.ai usage limit resets; turn it off in
/config("Continue automatically at usage limit")
It's on by default. Nothing needs enabling to get it; the toggle in /config is there for turning it off, not on.
What the limit messages actually mean
Claude Code's error reference documents three variants of this message, and they mean different things:
You've hit your session limit · resets 3:45pm
You've hit your weekly limit · resets Mon 12:00am
You've hit your Opus limit · resets 3:45pm
Subscription plans (Pro, Max, Team, Enterprise) include a rolling usage allowance measured on two windows at once, per Claude Code's cost management documentation: a five-hour session window and a weekly window, both shared across Claude chat, Claude Code, and Cowork. Usage counts against both simultaneously, so a single burst of heavy activity can exhaust the weekly allowance before the shorter session window even resets. The Opus limit is narrower still: it applies only to Opus requests specifically, and switching to another model with /model restores access immediately, no reset required. That makes /model the faster fix whenever it's only the Opus ceiling you've hit, rather than waiting on any reset at all.
A community-requested feature
This wasn't a surprise addition. A tracked feature request on the anthropics/claude-code GitHub repository had asked for exactly this months earlier: a way for Claude Code to detect a rate-limit response, wait out the reset, and continue automatically rather than hard-stopping and losing whatever state the session was in. The request proposed several possible implementations, a distinct exit code for scripts to detect, a --wait-on-limit flag, or simply keeping the session paused rather than exiting outright. The shipped feature in v2.1.234 doesn't match any single one of those proposals exactly, but it lands on the same underlying goal: a session that hits a limit shouldn't need a person to notice and manually restart it once the window reopens.
The /config toggle
The setting the changelog refers to lives under Continue automatically at usage limit in /config. Switch it off there to return to the previous behaviour, a hard stop with no automatic follow-up. There's no separate flag or environment variable documented for this specifically; /config is the interface.
The Desktop app's version of this
Claude Code's error reference also documents a Desktop-app-specific surface for the same idea, with more granular mechanics than the general changelog entry describes. In the Code tab of the Claude Desktop app, the card shown when you hit a session limit includes its own checkbox:
Auto-continue when limits reset: when checked, the Desktop app retries the interrupted turn after the reset and shows the retry time on the card. Uncheck it to turn this off for your account.
Two details worth knowing if you're on the Desktop app specifically:
- It resends the interrupted turn, not a summary or a fresh prompt. Whatever your last message was when the limit hit is what runs again once the window reopens.
- The weekly-limit card doesn't offer the same checkbox. Auto-continue in the Desktop app applies to the session-limit card only; a weekly limit still requires you to come back once the reset time documented in the message has passed.
Whether this Desktop-specific checkbox and the CLI-wide /config setting are the same underlying toggle surfaced two different ways, or two related but separately implemented mechanisms, isn't spelled out in either page. What's confirmed directly from Anthropic's own documentation is that both exist, both are described as retrying automatically once a limit resets, and the Desktop app's card is the more detailed of the two documented descriptions of what actually happens on retry.
The two windows, in more detail
Understanding what actually resets is worth a moment, because "session limit" and "weekly limit" aren't the same ceiling measured two ways, they're two independent windows you draw from simultaneously. Per Claude Code's cost management documentation, each Claude for Teams or Enterprise member's usage draws from a per-seat allowance sized by their seat tier (Standard or Premium), and that allowance resets on both a rolling five-hour window and a separate weekly window at the same time. The same pooled allowance covers Claude chat, Claude Code, and Cowork together, so heavy use of one counts against what the others have left.
That simultaneity is why a burst of unusually heavy work can exhaust the weekly window before the shorter session window has even come back around once. When that happens, auto-resume waits out whichever window is actually still closed, the message shown at the time of the hit names which one, session, weekly, or the narrower Opus-only ceiling, and the reset time attached to it.
If auto-resume isn't enough: usage credits
Auto-resume solves the waiting, not the ceiling itself. If you're hitting a limit often enough that the automatic wait is itself the bottleneck, the actual lever is usage credits, which let you keep working past your plan's allowance rather than stopping for the window to reset at all. Run /usage-credits after signing in through /login to manage them; what it opens depends on your role:
| Your role | What /usage-credits opens |
|---|---|
| Pro or Max subscriber | Settings > Usage on claude.ai, where you turn credits on or off and see your balance and spend limit |
| Team or Enterprise member with billing access | Your organisation's admin usage settings |
| Team or Enterprise member without billing access | A confirmation, then a request sent to your org's admins |
Usage credits and auto-resume solve genuinely different problems and pair well together: credits raise or remove the ceiling itself for anyone willing to pay for extra usage, while auto-resume makes the ordinary reset wait invisible for anyone who's fine working within the plan's normal allowance and just doesn't want to babysit the clock.
Why this matters for anything unattended
Anything you've set up to run without you watching it directly benefits most from this landing. A scheduled task, a long /loop, or a goal that keeps checking on background work all used to have the same failure mode: hit a session limit at 2am, and the work simply stopped until a person noticed and restarted it by hand. With auto-continue on by default, a task that runs into a session-limit wall now has a real chance of finishing on its own once the window reopens, rather than silently stalling until the next time someone happens to check in.
That said, this is specifically about the session or usage limit stopping a turn. It's unrelated to auto-compaction, which summarises a long conversation as it approaches the context window's limit rather than Claude Code's plan-based usage allowance; Claude Code's own cost documentation is explicit that a context or auto-compact warning is "not a usage limit" and shouldn't be confused with one.
The three limit types, side by side
| Limit | What it covers | Resets on | Fastest way past it |
|---|---|---|---|
| Session limit | A rolling five-hour usage window | The time shown in the message | Wait, or auto-resume if enabled |
| Weekly limit | A rolling weekly usage window | The time shown in the message | Wait, or auto-resume if enabled |
| Opus limit | Opus-model requests specifically | The time shown in the message | Switch models immediately with /model |
The first two share the same underlying auto-resume behaviour described above. The Opus limit is functionally different: because it's scoped to one model rather than the account's whole allowance, running /model and picking Sonnet or another available model restores access right away, without waiting on any reset at all, automatic or otherwise. Worth remembering if a scheduled task or an unattended session hits the Opus ceiling specifically rather than the broader session or weekly one, since auto-resume waiting on a reset is strictly slower than a model switch in that one case.
What it doesn't do
Auto-continue resumes a session, not a plan. If you're consistently landing on the session or weekly ceiling, the underlying fix is either a plan change or usage credits, which let you keep working past the allowance rather than waiting on any reset, automatic or otherwise. Auto-continue makes the wait itself invisible when there's nothing else to do about it; it doesn't raise the ceiling you're hitting.
It also doesn't override an Opus-specific limit by switching models for you. If the message names the Opus limit specifically rather than the session or weekly one, running /model yourself and picking a different model is still the immediate path back to working, faster than any reset.
A worked example
Say a scheduled task kicks off overnight, working through a long migration that ends up needing more turns than the current session window allows. Before v2.1.234, that task would hit the wall, print the reset time, and simply stop, leaving whatever state it was in until a person opened a terminal the next morning and manually continued it. With auto-continue on, the same task instead waits out the reset on its own and picks the interrupted turn back up, so by the time anyone checks in, either the task has finished or it's made real further progress rather than having sat idle the whole time.
Troubleshooting
The session isn't resuming after the reset time passed. Confirm you're on Claude Code v2.1.234 or later with claude --version, and check /config to see whether Continue automatically at usage limit is actually turned on; it's on by default from that version but can be switched off.
The Desktop app resent an old message I didn't want repeated. That's the documented behaviour: the Desktop app's auto-continue specifically retries the interrupted turn, not a fresh one. Uncheck Auto-continue when limits reset on the session-limit card if you'd rather review and re-send manually after a limit hit.
A weekly-limit hit isn't auto-continuing in the Desktop app. That's expected per Anthropic's own documentation: the auto-continue checkbox is offered on the session-limit card only, not the weekly-limit card. A weekly limit still needs you to come back after the reset shown in the message.
I want the old hard-stop behaviour everywhere. Turn off Continue automatically at usage limit in /config. There's no separate environment variable documented for this; the settings menu is the interface for both enabling and disabling it.
Where to go next
For the fuller picture of how Claude Code's usage windows, spend limits, and usage credits work together, see Claude Code's own cost management documentation. For running tasks unattended in the first place, where this change matters most, see Claude Code Routines vs /loop and what Claude Code's Monitor tool does. Browse the current Claude Code catalogue at getclaudeskills.com/platforms/claude-code.
Verified 22 August 2026 directly against Claude Code's own changelog, error reference, and cost management documentation at code.claude.com.
