← Back to digest
Commentary

Two model IDs for one model

What's this about?

dotey noticed that `claude-fable-5-1[1m]` and `claude-fable-5-1` appear to be two different model IDs, and asks whether they are actually different models.

Why would a context window get its own model ID?

Because a 1M-context variant is often a different deployment rather than a flag: different serving infrastructure, potentially different attention handling, and certainly different pricing. Exposing it as a distinct ID is honest, and it also means "which model am I using" has an answer more specific than the name on the box. dotey's finding yesterday that disabling the 1M context makes tokens last noticeably longer is the practical consequence — the two IDs behave differently enough to matter to your bill.

Why is it worth noticing?

24 likes
  1. If you are pinning model IDs in production code, the bracket suffix is not cosmetic.
  2. It connects directly to his `CLAUDE_CODE_DISABLE_1M_CONTEXT` finding — that setting is choosing between these.
  3. Small, precise observations like this are how the practical documentation of these products gets written.
← Ten months of agentsPrototype first, in HTML →