Cut your AI cost IN HALF (EASY)
TL;DR
Planning needs frontier models, execution doesn't: Use your best model for research and architecture, then hand off to a cheaper coding model for implementation.
A detailed spec bridges the capability gap: Berman shows a 600-line spec written by Claude Fable that cheaper models can execute effectively without frontier-level reasoning.
The math shows 68% savings: A feature costing $9.50 with Fable alone drops to $3.02 when coding is routed to a cheaper model.
Third-party tools auto-route because it differentiates them: Cursor, Factory, and Devin invest in model routing while frontier labs like OpenAI and Anthropic have no incentive to help you use less of their premium models.
Thinking levels are a hidden cost lever: Simple tasks like deployments don't need max thinking effort, and defaults are often higher than necessary.
Coinbase proves this at enterprise scale: Their token usage climbs while costs stay flat by routing to appropriate models and integrating open-source options like GLM 5.2.
Summary
Model routing, the practice of using frontier models only for planning and cheaper models for execution, can cut your AI costs by nearly 70%. Matthew Berman walks through the math, workflow, and tools to implement this strategy across manual setups and automated platforms like Cursor.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.