claude-haiku-4-5 is a 400 on Sume: the dotted id, and Haiku 5.5 news

Anthropic's alias for the model is claude-haiku-4-5, and its dated id is claude-haiku-4-5-20251001. Neither is a Sume catalog id. Sume's Haiku row is spelled with a dot, claude-haiku-4.5, and it is available only where the workspace catalog lists it. A Format run that sends the dashed alias or the dated id as model gets 400 invalid_request.

Separately, Anthropic's page lists Haiku 4.5 retiring no sooner than October 15, 2026, and its Sonnet 5.5 announcement says Claude Haiku 5.5 will join the 5.5 family in the coming weeks.

What does the vendor list for Haiku 4.5?

The models overview gives the facts below. Haiku is the one current model there without the 1M context window.

Which spelling does Sume accept?

In the repository's catalog code, Haiku 4.5 appears as anthropic/claude-haiku-4.5 with the spelling claude-haiku-4.5. I called the resolver for each variant. The dotted spelling resolved when the extended catalog was open and failed when it was closed. The dashed alias, the dated id, haiku-4.5 and claude-haiku-5-5 all failed in both cases.

This is a gated row, so the first thing to check on any workspace is whether its catalog lists Haiku at all. The call doc says an id outside the catalog is a 400 invalid_request, which is what a closed gate looks like too.

Is Haiku 5.5 something to wait for?

Anthropic's Sonnet 5.5 page says Claude Haiku 5.5 will join the Claude 5.5 family in the coming weeks. It gives no date, price, context window or API id, so none of those can be planned around yet. A Sume catalog row for it does not exist in the repository today either, and a request for claude-haiku-5-5 is a 400.

If a Format uses Haiku 4.5 for cheap steps, the retirement date is the practical deadline for a direct Anthropic integration. For Sume, watch the catalog and the echoed model.

What to do now

A 400 here costs nothing but a retry, so test the id with one call first.

• Replace claude-haiku-4-5 with claude-haiku-4.5 only if your workspace catalog lists Haiku.
• Otherwise omit model, or pick a row the catalog does list, such as claude-opus-5-5.
• Re-check the Anthropic overview after Haiku 5.5 ships.

Why does the spelling differ at all?

A vendor alias and a catalog id do different jobs. Anthropic's alias is a moving pointer to a model snapshot on its own API; a Sume catalog id names a row in the platform's registry, with its own routing and gate. The registry keeps a list of accepted spellings per row, and the dashed alias is not on the Haiku row's list.

The practical lesson matches the other models: copy ids from the catalog your workspace shows, and treat a 400 on model as a sign the id is not a row there, not as an outage.

The cost side is simple to read from the vendor page: at $1 input and $5 output per 1M tokens, Haiku 4.5 is half of Sonnet 5.5's $2 and $10, but it has a 200K window against 1M and no adaptive-thinking default. Whether that trade suits a given step is something to test on your own input, not assume from the price.

How do I find which Haiku my workspace can use?

Send a small test call with the dotted id and read the response. A 400 invalid_request that names the model field means the row is not in your catalog, and a run that starts and echoes the id back means it is. Do this once per workspace and save the result with the call, so a later 400 points you at a catalog change rather than a typo.

The Format call doc describes the field and the error; the Anthropic pages describe the vendor side. Neither tells you what a particular workspace lists, which is why the test call is the only reliable check.