What a session cost (/cost)

/cost breaks the session down by model:

• tokens in and out
• how many input tokens were served from the prompt cache
• the dollar cost, with a total when more than one model billed

Subagents and reviews often run on a different (cheaper) model, so they're listed separately rather than folded into the main one. In the desktop app, the breakdown also lists what each subagent spent (already counted in its model's row).

Watch the cached count on a long session. Those tokens are real context, but they bill far below the base input rate, so a big number there is why a long conversation costs less than its token count suggests.