Skip to content
← Tools
AI & Agentic Systems

Agent cost & context calculator

An agent loop re-sends its whole conversation every turn, so cost grows with the square of turn count. See what your loop actually costs, and what caching changes.

List prices as published, June 2026. Editable below — check against current pricing before quoting anything.

8.0K

The stable prefix, sent on every single turn. This is what caching is for.

3.0K

User message plus tool results. Tool results are usually the larger half.

800

What the model generates, including its thinking.

12

The agent loop depth. This is the number that changes the arithmetic.

500

How many times this runs across the business.

Per session

$1.97

347K in · 9.6K out

Per day

$987

500 sessions

Per month

$29,610

30 days at this volume

What you are paying to send

  • 3% new content
  • 97% re-sent history

Without caching

$29,610

per month

With caching

$11,166

per month · 62% less

Context at the final turn

54K of 1.00M

Critical

97% of input tokens are re-sent conversation history

The API is stateless, so every turn re-sends everything before it. Across 12 turns that is 347K of input for 11K of genuinely new content on turn one. Input cost grows with the square of turn count, not linearly — which is why an agent that is affordable in a three-turn demo is a different product at 12.

Worth knowing

Prompt caching is off

Enabling it would take this session from $1.97 to $0.74 — 62% less. Caching is a prefix match: any byte that changes early invalidates everything after it, so a timestamp or a per-request ID near the top of the system prompt silently disables it and nothing errors.

Prices are list prices as published in June 2026 and are editable above — they will drift, and this page is not a price list. The arithmetic is the durable part: input cost grows with the square of turn count because the API is stateless and every turn re-sends the whole conversation. That is the shape most agent budgets get wrong, at any price.