Long support histories open reliably
Long support requests now open reliably: message and status history is split into pages and no longer runs into Telegram's message-length limit.
Hotfix¶
Anthropic cache prices are calculated correctly
We corrected the price calculation for POST /v1/messages with Anthropic
prompt caching so response usage, history, and the final debit agree.
Model availability and prices display correctly
This urgent fix synchronizes the API, Telegram bot, and documentation catalog so you see only models and prices that your key can actually use.
Claude compatibility and model access restored
This urgent correction restored requests after the previous update and synchronized the available-model view in the API and Telegram bot.