Products vs models — the distinction
subagentmodels.com catalogs Claude model names -- claude-sonnet-5, claude-fable-5, claude-opus-4.8, and so on: the actual inference engine that generates a response. This site, subagentproducts.com, catalogs Claude product surfaces -- Claude.ai, Claude Code, Claude Cowork, and so on: the interface a person or developer uses to reach whichever model is currently serving that surface.
These are orthogonal axes, not a hierarchy with one nested inside the other:
- One surface, many possible models over time. Claude.ai has been served by many different model generations across its lifetime; the surface (a chat interface) stayed conceptually the same while the model backing it changed.
- One model, many possible surfaces. The same underlying model release can answer requests from Claude.ai, from Claude Code, and from a raw API call -- the model doesn't know or care which surface originated the request.
- Surfaces can layer on other surfaces. Claude Cowork is a collaborative-workspace mode that itself runs on top of the same session infrastructure as Claude Code / Claude Desktop, rather than being a wholly separate product from first principles.
Concretely in this repo: crates/schema/src/agent.rs's AgentModel enum (cataloged on subagentmodels.com) describes which model a generated sub-agent runs on. It says nothing about which product surface invoked that agent -- that's a separate fact, and exactly the kind of row this site exists to hold instead.
See also: what a product surface is, and subagentmodels.com's own README for its deliberately narrow scope (only models this repo's own code actually depends on).