The Author in the Machine // the bill

What did that conversation just cost?

Type in the token counts from the agent run. Same conversation, three brains — watch what happens when the author gets more expensive.

input tokens (everything the model read)

prompt + tools + history, every turn, added up

output tokens (everything it wrote)

the replies + its tool-calling chatter

or try a known size:

the three documents assume the model reads the whole thing and writes back a ~1-page summary.

Gemini 2.5 Flash

$0.30 in · $2.50 out / 1M tokens

—

what our demo agent runs on

Claude Opus 4.8

$5.00 in · $25.00 out / 1M tokens

—

Claude Fable 5

$10.00 in · $50.00 out / 1M tokens

—

now scale it

people doing this

≈ Lloyds Group headcount

conversations / person / day

working days / year

Prices are list API prices per million tokens (June 2026): Gemini 2.5 Flash $0.30/$2.50 · Claude Opus 4.8 $5/$25 · Claude Fable 5 $10/$50. Caching, batching and enterprise deals change the real bill — the point is the shape, not the pennies.