"Which model should we use?" is the wrong question. The right one is "which model should handle this call?" — and the answer changes with the workload, the latency budget, and what you're willing to pay to be wrong less often.
These posts are about making that call deliberately: small-first escalation, benchmark skepticism, and treating model choice as a routing decision instead of a one-time procurement.
'Which model do we standardize on' is procurement thinking applied to a per-call decision. Escalation rules, premium criteria, and eval gates are one framework wearing three hats.
The top-tier model is worth 10x on maybe a tenth of your calls. The trick is knowing which tenth — and it starts with what a wrong answer costs you.
Everyone prices the model ladder from cheap-API up. There's a rung below that: weights on hardware you already own, at a marginal cost of zero.
A model's leaderboard position tells you how it does on the leaderboard's distribution, not yours. Fifty queries from your own logs beat it every time.
Autocomplete, overnight reports, and research tasks live in different corners of the same triangle. Products that pick one corner for everything overpay in all three.
Most teams discover their switching cost the day a price hike or outage forces the move. Ask these six questions while the answer is still cheap.
Escalation isn't a cost hack, it's an architecture: the cheap model handles the bulk, the expensive one earns its rung. The same ladder logic that works for tool calls.
Model selection for agents isn't a leaderboard question. It's five workload questions, and most teams skip straight to the logo.