Everything you can call through LIBRA. Sign in to see your prices.
Send model libra/auto and LIBRA picks the best available model for each request: a vision model for images, a tool-capable model for tools, a strong model for code and reasoning, a fast one for short chat, and ASIX's local models for Indonesian. If that model fails, the next best one answers. Later turns of a conversation stay on the same model, and libra/auto:code (or :reasoning, :summarize, :translate, :extract-json, :chat, :chat-id) names the task when you already know it.
Billed at the price of the model that answered. Usage and Logs show which model that was.