Cheaper tokens for Claude Code

Updated 2026-11-08 · Use-case guide

Claude Code is one of the most token-hungry tools a developer can run: every file it reads, every plan it writes, every test it reruns is billed. Developers routinely report hundreds of dollars a month in usage on larger codebases.

Why the bill grows so fast

• Large context windows get re-sent on every turn
• Agentic loops retry and re-read files
• Frontier models carry the highest per-token price

Two ways to pay less

1. Reduce tokens: tighter CLAUDE.md instructions, smaller working sets, compacting often, delegating grunt work to a cheaper model.
2. Pay less per token: route usage through a discounted endpoint like Token Supply (up to 70% off, varies by model).

Most people do both. Our full cost-cutting guide covers the settings side; the discount side takes about two minutes.

Setup

Claude Code reads its endpoint and key from environment variables. After signing up, you'll set the base URL and key the vendor gives you in your shell profile, restart the terminal, and run as normal. Token Supply's dashboard has copy-paste instructions for macOS, Linux, and Windows.