← Back to PayloadsAI/ML2026-04-28
opus: The Intelligent Claude Routing Layer Built for Speed
opus uses a three-tier Haiku→Sonnet→Opus routing strategy to automatically pick the right model for each task — cutting costs without sacrificing capability.
Quick Access
Install command
$ mrt install ai

TL;DR
Intelligent Claude model routing with Haiku → Sonnet → Opus progression. The right model for every task at the lowest cost.
10-Second Pitch
- **What it does:** Classifies prompts by complexity, routes to the cheapest capable model
- **Key win:** Most tasks don't need Opus — opus makes sure you only pay for what you use
- **Best for:** Production Claude deployments where cost matters
Setup
pip install opus-router
opus route "What's the capital of Japan?" # → Haiku (trivial)
opus route "Debug this race condition" # → Sonnet (medium)
opus route "Design a consensus algorithm" # → Opus (complex)
Pros / Cons
| Pros | Cons |
|---|
| Automatic cost optimization | Classification isn't perfect |
| Preserves quality on hard tasks | Still need human review for critical outputs |
|---|
| Simple drop-in installation |
|---|
Verdict
opus is the routing layer your Claude deployment was missing. The difference between Haiku and Opus is 100x in cost — and opus makes sure you only pay premium prices when you actually need them.