Model Routing
Simple Definition
Model routing is the practice of sending different parts of an AI workflow to different models based on cost, speed, quality, or task type.
Not every task needs your strongest model. A smart workflow may use one model for planning, another for execution, and another for review.
Why It Matters
Routing helps you spend expensive, high-quality models only where they add the most value, while cheaper or faster models handle the routine work. It can lower cost and reduce reliance on any single model.
Example
A developer uses Claude for planning, Codex for implementation, and Claude again for the final review, matching each model to the part of the job it does best.
A team calling OpenAI’s API might send most support-ticket summaries to GPT-6.2 Sol at $2 per million input tokens and escalate only the tickets Sol flags as ambiguous to GPT-6.2 Astra at $8. If the summaries can wait until morning, running them in an eight-hour background window cuts the token rate by another 40%. See GPT-6.2 Astra vs GPT-6.2 Sol.
Routing vs. Orchestration
Strictly, routing picks which model handles a request. Orchestration goes further and has several models work on the same request: one drafts, a check decides whether the draft is good enough, and a stronger model takes over or another model reviews it. Products now do both automatically. Google’s Orchestrator in Antigravity, launched in November 2026, switches between a planning model (Gemini 4 Carbon, Argon, or Neon) and the faster Gemini 4 Hydrogen within one coding session, handing control back to the planner when Hydrogen gets stuck; see Gemini. Microsoft’s Decision 1.1, generally available in Microsoft Foundry since November 17, 2026, is a small, cheap model sold specifically to make routing calls inside your own app (see decision model). GitHub Copilot Fusion chooses per prompt between a single model, a cascade, and a critique. For the four levels from manual selection to evaluation-driven orchestration, and how to compare them on cost per completed task, see from routing to orchestration.
Related Terms
- Model Fallback, routing to a backup when the main model fails
- AI Workflow, the multi-step process routing optimizes
- AI Automation, automated systems often route between models
- Reasoning Model, typically used for the planning and review steps
- AI Coding Agent, coding workflows are a common place to route work
- Evaluation-Gated Routing, routing decisions backed by actual eval results instead of intuition
- Model Escalation, moving a specific task to a stronger model mid-workflow
- Decision Model, a bounded-output model that can make routing choices
- Related guide: AI token costs explained
Continue learning
Explore related guides, tools, workflows, and prompts that help you go deeper into this topic.
Browse all AI terms.
Learn termSee these concepts in practice.
Open workflowA simple explanation of this AI concept.
Learn termA simple explanation of this AI concept.
Learn termA simple explanation of this AI concept.
Learn termA simple explanation of this AI concept.
Learn termSee AI terms in action
Browse practical AI workflows that use the concepts in this glossary.
Last updated: