Why did that Claude Code or Codex session cost so much?

Claude Code’s /usage shows the current session’s cost, but not a ranking of past sessions. ACTVT estimates what every Claude Code and Codex session would cost at API rates, from its tokens and the model’s published prices. Sort the session list by cost to find the expensive ones, then open one. The inspector shows its cost, its tokens and cache hit rate, and how much of the total went to subagents. It’s an estimate at API rates, so if you’re on a plan, it isn’t your bill. The session browser and its sort are part of the Premium tier.

0:445 chapters

Step by step

  1. 0:00Two runs, the same job

    Two agents ran the same code review earlier, and I want to know if one of them cost more.

  2. 0:05Sort by cost

    I sort today’s runs by cost. The one that split the work across subagents is at the top.

  3. 0:15What it cost

    Here’s what it cost at API rates. If you’re on a plan, that’s not what you’re billed.

  4. 0:23Where it went

    Most of its input tokens were read from cache. It used six subagents, and $0.29 of the $0.52 was theirs.

  5. 0:34The same job, one agent

    The same review done by a single agent cost $0.10. It had the same cache hit rate and no subagents, so it came to about a fifth of the cost.

Questions

Is it my bill?
No, it’s an estimate at API rates; on a plan, it isn’t what you pay.
What’s in the breakdown?
Tokens, cache hit rate and the share spent by subagents.

Try it on your own agents

ACTVT is free to start, and reads the session files Claude Code, Codex CLI and Pi already write. Nothing to configure.