Model family

Labs ship models as tiered families: a flagship for maximum capability, a balanced mid-tier, and a small fast tier, refreshed together across generations. The tiers share training lineage and behavior style, which makes moving between them relatively predictable compared with switching vendors.

Families create a pricing ladder, with roughly an order of magnitude separating the cheapest and most expensive tier per token. Since a new generation's mid-tier often matches the previous generation's flagship, staying current on family releases is itself a cost strategy.

Within-family routing is the lowest-friction optimization available: prompts, tooling, and integrations mostly carry over, so easy traffic can flow to the small tier while hard tasks retain the flagship.