Changelog

API and platform changes.

Billing moved to pure USD pass-through: model rates equal provider rate cards, zero markup, revenue only from the 10% top-up fee
Per-request bill detail added (input / cache / output cost breakdown plus request_id)
Added deepseek-flash, qwen-flash, qwen-plus and qwen-max; retired deepseek-coder
Upstream prompt-cache segments and DeepSeek peak windows are now billed separately