Research · August 19, 2026
Dynamic model routing under a cost ceiling
Task-appropriate models, priced before dispatch, bounded by a budget the enterprise sets.
Summary
A technical note on dynamic model routing: how Orcher decouples a directive from any single provider, scores candidate models against task requirements, prices each option before dispatch, and holds total spend inside a ceiling defined per directive.
Full text
The record
Routing is the difference between using models and being dependent on one.
Decoupling the directive
In Orcher a directive states what must be done and under whose authority. It does not name a provider. Routing selects the execution path at dispatch time from the set of models the enterprise has approved.
Scoring the task
Candidates are scored against the requirement: the reasoning depth the task genuinely needs, latency tolerance, context size, boundary constraints on where the data may go, and price. Most enterprise tasks do not need the most expensive available model, and routing is where that observation turns into money.
The ceiling
Every directive carries a budget. Prices are computed before dispatch, budget is decremented as work proceeds, and the ceiling stops execution rather than annotating an overrun after the fact. Escalation to a more capable model is a decision the policy allows or refuses, recorded either way.
What accumulates
Each routed directive leaves behind which path was chosen, why, what it cost, and whether the outcome held. That record is tenant-owned and compounds as Dipp Intelligence, so routing decisions improve on the enterprise's own evidence rather than on a vendor's averages.
Corrections welcome, in writing.
If a figure we published is wrong, we want the citation. Research correspondence goes straight to the founding team.