The $5 Agent Turn

Three composable levers that cut the token bill

Every agent turn re-sends a huge fixed context — and per-token billing makes that repetition expensive

Amplifier agents run on frontier models billed per token. The foundation instruction surface alone measures 27,025 est. tokens; add agent prompts and history and a turn easily carries 40k+ tokens — re-sent every single turn.

The first lever attacks exactly this repetition — and it's already on.

27,025
est. tokens: foundation root surface (bundle.md + context/*.md), measured 2026-02-10
40k+
tokens carried per turn once agent prompts + history are added
On by default

A cache read costs exactly 10% of a fresh input token — and it ships enabled

On Sonnet 4.5, cached input reads at $0.30/M versus $3.00/M fresh — exactly 10%. In the Anthropic provider, enable_prompt_caching defaults to True, so the durable savings lever is live with zero configuration.

Free by default — but naive usage still leaves money on the table.

$3.00/M
Sonnet 4.5 fresh input tokens
$0.30/M
Sonnet 4.5 cache read — exactly 10%
True
enable_prompt_caching default in the Anthropic provider
The Complication

Naive usage pays full input price for repeated context — and routes trivial classification to the most expensive model

Without caching applied, that 40k+ context is billed at full $3.00/M every turn. And sending simple parsing or classification work to a frontier model means paying top rates for tasks a cheap model handles fine.

Two more levers close that gap — here's how each one works.

Caching turns repeated context nearly free

The provider stamps cache_control on the system prompt, tools, and the last message. Re-sending a 40k context over 10 turns drops from ~$1.20 to ~$0.258 on input alone — about 78.5% saved on the repeated context.

That's the default lever. The second one is a single declaration away.

Naive input, 10 turns of 40k~$1.20
Cached input, 10 turns of 40k~$0.258

~78.5% saved on repeated context (illustrative model built from verified Sonnet 4.5 rates, not a measured bill).

Match cheap work to cheap models — Haiku, not Opus

Agents declare a semantic model_role. The 'fast' role — parsing, classification, file ops, bulk work — routes to Haiku instead of Opus, which is 5x cheaper per token on both input and output.

The third lever governs what happens once the loop is running.

$0.015
Opus: a 2k-in / 200-out classification ($5/M in, $25/M out)
$0.003
Haiku 4.5: the same call ($1/M in, $5/M out) — 5x cheaper

The loop bounds and governs runtime cost

Amplifier's streaming loop executes multiple tool calls concurrently with deterministic ordering. A reference cost-aware scheduler hook watches tool selection and can veto or swap an expensive choice against a cost budget.

Stack all three, and the bill changes shape entirely.

The Payoff

Stacked, the three levers collapse an illustrative ~$50 workflow toward ~$5

Caching alone saves ~78.5% on repeated context; routing cuts trivial work 5x; the loop bounds and governs the rest. Composed, the optimizations are order-of-magnitude — and the biggest lever, caching, needs zero configuration.

The ~$50 → ~$5 figure is an illustrative headline, not a measured bill — the defensible numbers are 10%, ~78.5%, and 5x.

Composable and defaults-first, not hand-tuned

Turn on nothing to get caching. Declare a model_role to route cheap work to cheap models. Let the loop govern the rest. Stack cheap-by-default levers rather than hand-tuning a bill.

The order-of-magnitude win comes from composition, not configuration.

Sources

Research Methodology

Data as of: July 21, 2026. Grounded in Amplifier repos under /home/ramparte/dev/ANext plus the primary-source cost-aware scheduler.

Feature status: Caching SHIPPED & on by default; routing-matrix SHIPPED; loop-streaming SHIPPED; cost-aware scheduler is a reference implementation.

Commands run:

Gaps: The ~$50 → ~$5 headline is a rounded, illustrative figure — NOT found verbatim in any repo. The 40k+/turn figure is inferred (measured value is the 27,025-token root surface). Scheduler tool_costs are example units, not USD.

Primary contributors: Brian Krabach (sole author of all 14 scheduler commits; top contributor with 38 commits to routing-matrix).

More Amplifier Stories