Official Claude Code Token Saving Guide: 11 Tips to Extend Your Quota

Anthropic has released the Claude Code Token Saving Guide. The core message is simple: the longer the context, the more expensive each subsequent round becomes. Chatting, document handling, and command outputs will continue to consume tokens, thus depleting your quota faster.

Here are 11 practical tips:

1. Use /clear to switch tasks. Clear the context after completing a task. If you are still working on the same task but the previous content is no longer useful, use /compact to compress it. Longer conversations become more expensive as each round will reintroduce previous content.

2. Choose the model and effort settings at the start. Executing /model or /effort midway will clear the prompt cache, requiring the entire context to be processed again in the next round. Fast mode should also be enabled from the beginning.

3. Compact before leaving. Subscription users have a cache that expires after about 1 hour, while the API Key defaults to about 5 minutes. If you return after the cache expires, the next round will usually require reprocessing the entire context. Compressing while the cache is still valid is cheaper.

4. Use /rewind if the last few rounds have gone off track. It only deletes the last few rounds, allowing you to continue using the earlier cache. /compact rewrites the entire conversation, which is more costly.

5. Use @filename directly for files. Claude will access the file in the first round, saving a Read or search step. A file typically only needs to be referenced with @ once; repeating @ may inadvertently add the same file back into the context.

6. Be specific in your prompts. Instead of saying "the test failed," directly tell Claude which test and which file. Otherwise, it may grep, search, and read through multiple files, and these results will continue to occupy context in each subsequent round.

7. Minimize output from tests and logs. Outputs from builds, tests, and git logs will remain in the context. You can add a quiet parameter to only output errors or use tail to truncate. Claude Code will automatically write command outputs exceeding 30,000 characters to a file, leaving only the summary and path in the session.

8. Use /context to check startup load. At the beginning of a new session, you can see how much context is occupied by CLAUDE.md, MCP, and tool definitions. CLAUDE.md should only contain general rules, while specialized processes should be loaded on demand through Skills, and unused MCPs can be turned off with /mcp.

9. Use smaller models and lower effort for simple tasks. Larger models are more expensive for both input and output, and thinking also counts as output tokens. For purely execution-based tasks, there’s no need to always use the highest-tier model and inference intensity.

10. Delegate large logs to subagents, and avoid misusing them for small tasks. Subagents have independent contexts and only bring back the final results to the main session, making them suitable for reading large logs or searching documents. However, they also need to read files again and run multiple rounds themselves, which can make small tasks more expensive. Tasks like searching and reading logs can also specify Haiku or Sonnet.

11. It’s best to run /loop in a separate session. Each loop is a complete round and will carry all the current session's context. If the interval exceeds the cache time, it may also accumulate a cache expiration. The official recommendation is to open a new terminal and run loops in a clean session.

Ultimately, Anthropic's advice boils down to: avoid stuffing useless content, clear tasks promptly, and don’t let sessions drag on too long.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.