MODEL Not measured · roadmap capability

Unit Economics Model

A back-of-envelope model for how tier routing could price out. Arithmetic on assumed rates, not observed spend.

Pricing Index: USD per 1M tokens

Configure Task

Task Type

Input Context Size (Tokens) 18,000

Includes codebase symbols, dependency trees, and standard layers.

Target Output Size (Tokens) 1,200

Estimated synthesized code blocks, patches, or verification logs.

Flat Frontier Cost

$0.0720

All tasks routed to Tier 3 Frontier API.

Tier-Routed Cost

$0.0084

Modelled hybrid execution.

Modelled Difference

88.4%

Arithmetic on assumed rates.

What this is and is not

Tier routing is planned work, not shipped work, and ContextStack has not measured its own spend against a frontier-only baseline — real-path token accounting is still being instrumented. Every number here is arithmetic on the assumed rates shown, with a modelled 30% context reduction from workspace isolation.

For what routing has actually achieved elsewhere, published third-party results are the honest reference point: Lindy reported roughly a 90% inference-cost reduction on migrated traffic, and the RouteLLM paper reports around an 85% cost reduction while retaining 95% of baseline quality. Neither is a ContextStack measurement.

Router Verdict

TIER2

Assigned Model Backend: Hosted open-weights model

Tier 2 covers mid-range reasoning. Routing a moderate refactor to a fast hosted open-weights model is where the modelled difference is largest — and also where it most needs verifying against real gate-pass rates before anyone should believe it.