Open Source vs Closed Source
Open-weight vs proprietary models: what open really means, the true cost of a token, and when each one wins.
Cost Per Task, Not Per Token
A model that’s half the price per token can cost the same per finished task, because the cheap one burns twice the tokens. Here’s the number to actually compare.
Tip #204Open Weights Isn’t Local, and Isn’t Free
The thumbnails say free, local, open source. Only one of those three is really true for a frontier open-weights model. Here’s what open weights actually buys you — and what it doesn’t.
Tip #205Plan Expensive, Build Cheap, Review Sharp
Don’t hand one coding task to one model. Split it across three: plan with your strongest model, execute with a cheap fast one, review with a second frontier model. Same result, roughly a third of the cost.
Tip #206Point Your Coding Agent at an Open Model
Your coding agent doesn’t care which model sits underneath. Point its env vars at a provider serving an open model and the harness never notices. Here’s how — and when to keep a frontier model one flag away.
Tip #207Caching Is Where Open Models Get Cheap
Open models look cheap per token — until the bill lands. Input caching is the setting that makes them actually cheap.
Tip #208Benchmarks Flatter New Models
A new open model “ties the frontier” on the charts, then takes about 90 minutes and 21M tokens to do what the leader did in 17. Benchmarks flatter new models — run your own hardest task before you switch.