Quick answer
Chat and Claude Code share the same session and weekly windows. Claude Code usually consumes more because it resends a large repository context on every turn, while chat sends a shorter conversation. Use the account percentage alongside local Code logs to see which side moved.
Why Code costs more
| Factor | Chat | Claude Code |
|---|---|---|
| Context | Conversation | Repository plus conversation |
| Turns | A few | Many per task |
| Tool output | Rare | Frequent |
| Caching | Moderate benefit | Large benefit when stable |
| Typical share | Lower | Higher |
Attribute the drain
- Note the weekly percentage at the start of the day.
- Use only chat for a few hours and note the change.
- Use only Code for a comparable task and note the change.
- Compare the two deltas.
- Adjust the heavier side first.
Act on the result
- If Code dominates, shorten sessions and trim instructions.
- If chat dominates, start new conversations and avoid huge attachments.
- If both are low but the limit still falls, another device is using the account.
Why chat and code spend differently
Both count against the same windows, but they consume differently. A chat turn carries a short conversation; a coding turn carries files, tool output, and a longer instruction set. Per minute of attention, coding is far more expensive.
- Chat is cheap per turn and rarely the cause of a block.
- Coding agents multiply context with every tool call.
- Desktop and web usage are invisible locally but still counted.
- Weekly caps are usually spent by coding, not chatting.
The desktop tracker guide covers the blind spot, and UsageAtlas tracks the coding half accurately.
Questions, answered
Does Claude Code use more of my limit than chat?
Usually yes, because of repository context and repeated turns. The exact ratio depends on how you work.
Are the limits shared across devices?
Yes. The allowance belongs to the account, not the device.
Can I limit Claude Code separately?
Some organizations set per-seat policy. Individual plans share one allowance across surfaces.