Looks like it's about 4M input tokens or 0.5M output tokens with their cheapest model `@cf/meta/llama-3.2-1b-instruct`

With `@cf/qwen/qwen3.8-27b` it's 240K input tokens and 34,000 output tokens