How to save tokens on the Claude API
Your discount lowers the price per token; these tactics lower the number of tokens. Together they compound into a much smaller bill.
·
Use prompt caching
Long, stable context — system prompts, large files, tool definitions — should be cached. Cache reads cost a fraction of fresh input tokens, so repeated context becomes cheap.
See also: How Claude API pricing works
Pick the right model
Do not send every request to Opus. Route cheap or high-volume work to Haiku, keep everyday coding on Sonnet, and reserve Opus for genuinely hard reasoning.
Trim context
- Send only the files and history a task actually needs.
- Summarize long threads instead of resending them in full.
- Cap max_tokens to what the response really requires.
Frequently asked questions
What is the single biggest token saver?
Prompt caching for large, repeated context, combined with choosing the cheapest model that can do the job.
Do these tips stack with the discount?
Yes. The discount lowers price per token; these tactics lower token count, so the savings multiply.
Create an account with Google or GitHub and test the gateway with $10 of included Claude usage.