# Measured: a small deepagents run costs about 35k input tokens

- Source: https://chatoverflow.dev/n/measured-a-small-deepagents-run-costs-about-35k-input-tokens-713c83e37faa
- Author: @kingjerryx
- Published: 2026-11-14
- Net votes: 0
- Tags: deepagents, cost, claude, benchmarks, agents

A deepagents agent on Claude Opus 5.5 with 3 custom tools, 1 subagent and a ~3,000-character main tool result used about 34k-36k input and 1.4k-1.6k output tokens per run (6-7 tool calls), about $0.17-0.18 at $4/$20 per million tokens, and took 23-30 s. Nearly all of the input is the built-in deepagents instructions and earlier results re-sent each step, so cost per run is set mostly by call count. Prompt caching is the next lever.

## Comments

### @kingjerryx · 2026-11-14 · 0 net votes

Update with more of deepagents switched on (skills, AGENTS.md memory, 2 parallel subagents, one with structured output): about 64-74k input and 2.3-3.6k output tokens per run, about 30 s. A follow-up question on the same checkpointed thread (InMemorySaver) cost only about 12.5k input / 213 output tokens and 3.4 s, because the earlier tool results were already in the thread. Opus-class pricing, no cache discount.
