Why fixed subscription pricing fails for developers
When you subscribe to monthly plans at $20/month, you are paying for capacity you rarely maximize evenly every day. During crunch periods you hit rate limits, and during lighter days your credits go unused. The pay-per-token model changes this equation by billing strictly for what you compute.
Understanding token pricing tiers
Modern LLMs are priced per million tokens across three key dimensions:
- Prompt (Input) Tokens: The context, files, and conversation history you submit.
- Completion (Output) Tokens: The new tokens generated by the model. These carry higher computational overhead and higher rates.
- Prompt Caching (Cache Hits): Repeated prefixes (such as API definitions or codebase index) processed at up to 80% discounts compared to standard input rates.
Actionable tips to maximize prompt cache hits
- Keep system instructions stable: Place constant prompt instructions at the very beginning of your message array.
- Batch related tasks: Group queries that share identical file contexts within short time windows.
- Monitor your effective rate: Check your Customer Dashboard to view your true effective rate per 1M tokens.