Models
What is available and when to reach for which.
Every model is served through Anthropic's API. Requests are proxied so the treasury can meter spend; nothing about the model's behaviour is modified.
Haiku 4.5#
Fast and cheap, 200k context. The right default for quick questions, drafting, and anything conversational. At 1.5 credits a request, a small quota still buys real usage.
Sonnet 5#
The workhorse, 1M context. Use it for real writing, reading long documents, and multi-step reasoning where Haiku starts guessing. Three times the cost of Haiku.
Opus 5#
Heavy agentic work, 1M context. Large codebases, long research, anything that needs many steps held together at once. Holders only — not because guests are untrusted, but because a free tier funding 7.5-credit requests would drain the pool in an afternoon.
Picking a bigger model does not get you a bigger quota. It gets you fewer, better requests out of the same one.