/// signals/#002
Why Opus 4.7 over Sonnet for the chat agent.
Latency cost we're happy to pay for fewer recovered conversations.
Sonnet 4.6 is faster and cheaper per token. Opus 4.7 is roughly 3× the cost.
We pick Opus on the assistant because every recovered conversation it produces - the kind where it answers the question instead of saying "I'd need more context" - pays for ~40 Sonnet sessions in the same surface. The tool-use loop also tightens: Opus reaches the right tool in 1 turn where Sonnet sometimes needs 2.
When latency matters more than reasoning depth (transactional UIs, classification), it flips the other way. Match the model to the failure mode you can least afford.