Claude Cowork usage limits — what to do when you hit the wall

TL;DR. Cowork shares Claude's usage envelope; limits reset on a rolling 5-hour window. Pro gets roughly 1–1.5 hours of intensive Cowork work per window; Max gets ~5× that; Team and Enterprise have org-level pools. The mid-market team that hits limits hardest is usually one power user on Pro with four light users. Moving the power user to Max is usually the fix.

How limits work#

• Cowork shares Claude's usage envelope. Limits reset on a rolling 5-hour window.
• Heavier models consume faster. Opus runs through the envelope much faster than Sonnet; Sonnet faster than Haiku.
• Long-context runs and sub-agent runs consume more than equivalent work in smaller chunks.
• Pro — ~1–1.5 hours of intensive Cowork work per 5-hour window.
• Max — ~5× the Pro envelope.
• Team / Enterprise — org-level limits, configurable.

Verify the specific numbers against current Anthropic disclosures — these change.

What to do this hour#

• Wait for reset. The timer is shown in the UI.
• Switch to a lighter model if mid-task. Sonnet 4.6 → Haiku for the simpler bits.
• Break the work into smaller chunks. Sub-agent runs share the same envelope, but you can spread them across the reset.
• Pause and switch to non-Cowork work. Come back after the reset.

Pre-empting the wall#

• Use Sonnet 4.6 as the default model. See Models and context.
• Pre-summarise long inputs into a leaner prompt context — chunk before processing.
• Don't run Opus on file ops. The cost math doesn't pay.
• Schedule heavy tasks for off-hours so the limit hits when you are not waiting on the result.

Right-sizing your plan#

The "user is on the wrong plan" signal#

• Operator stops using Cowork mid-week because of limits.
• Operator works around limits by sub-optimising prompts (using simpler-than-needed models, skipping plan steps).
• Manager notices throughput dip on Wednesdays after Tuesday-heavy use.

These are upgrade signals, not training problems.

Coordinating across a team#

• Don't run scheduled tasks and operator work on the same seat. Schedule consumes the operator's quota.
• Concentrate scheduled tasks on a dedicated Max seat.
• For Team / Enterprise, monitor org-level usage in the admin console monthly.

Tinkso's take#

The mid-market team that hits limits hardest is usually one with a single power-user operator on Pro and four light users. Move the power user to Max and the team's effective throughput jumps without changing headcount. The Pro→Max upgrade pays back in less than a month for daily users.