Cursor’s flat $0.25/M token surcharge on Luna turns a $0.02 cache read into a $0.27 charge — a 12.5x markup on the cheapest token type in the stack.
The breakdown against OpenAI’s base rates, per an Enterprise Cursor user on Reddit:
- Input tokens: $0.20 base + $0.25 surcharge = 125% markup
- Output tokens: $1.20 base + $0.25 surcharge = ~21% markup
- Cache reads: $0.02 base + $0.25 surcharge = 1250% markup
The issue is structural. Cursor applies one flat surcharge across all token types. Coding workflows generate 90%+ of token volume as cache reads — the token type designed to be cheapest, at roughly 1/10th of input cost. A flat surcharge inverts that economics. The same pattern hits GPT Sol, where $0.50 cache reads carry the same $0.25 surcharge — a 50% markup, less extreme but still disproportionate.
NOTE
LLM API costs consume roughly 70% of SaaS gross margin for AI coding tools, per analysis on Medium. Cursor’s surcharge bridges that gap — but flat markups on cache reads target the wrong token type.
Enterprise users running long agent sessions should audit cache-read volume before committing to Luna. A session generating 10M cache reads costs $0.20 at OAI rates and $2.70 through Cursor — and that gap scales linearly with usage.