# Claude run receipts and accounting caveats

Recorded August 10, 2026.

| Run | Reported duration | Reported agent tokens |
| --- | ---: | ---: |
| Claude Fable 5, run 1 | About 12.0 minutes | Unavailable as a clean per-run number |
| Claude Fable 5, run 2 | 13.5 minutes | 128,497 |
| Claude Opus 5, run 1 | 21.7 minutes | 158,284 |
| Claude Opus 5, run 2 | 18.3 minutes | 146,001 |
| Claude Sonnet 5, run 1 | 13.9 minutes | 131,523 |
| Claude Sonnet 5, run 2 | 14.9 minutes | 132,496 |

These figures were supplied by the Claude benchmark operator after the runs completed.

## What they do not prove

- The six Claude builds ran concurrently on one machine. Their reported durations are not a controlled speed ranking.
- The `agent tokens` field is not broken into fresh input, cached input, and output. It is not equivalent to the Codex receipt fields.
- Fable run 1's parent session included earlier setup and orchestration work. Its aggregate `/cost` response was excluded instead of being attributed to the build.
- These are not itemized Claude Code charges. We do not calculate API-equivalent job costs from them.

Current API prices can provide sticker-price context, but they cannot repair missing job receipts. Anthropic lists Fable 5 at $10 per million input tokens and $50 per million output tokens, Opus 5 at $5 and $25, and Sonnet 5 at a temporary $2 and $10 through August 31, 2026 before its standard $3 and $15 rates.

- [Anthropic model overview and current pricing](https://platform.claude.com/docs/en/about-claude/models/overview)