Ultracode finds the right model id on your own key

When you switch on Ultracode in a chat, the agent can start a workflow with several models. It has to name the model for the run. With your own Anthropic key the real ids carry a prefix, for example anthropic-byok/claude-opus-4-8. The agent did not know that. It passed the bare name you had used in the chat, and the platform answered that the model was not available on your account.

The tool description now shows the model of the current chat and the ids your account may use. A bare name such as claude-haiku-5-5 is matched to your account, and the provider of the current chat wins when two ids fit. If the name matches nothing or several ids, the answer lists the valid ids, closest first, at most twenty, so the agent can correct itself. The same check runs when you start Ultracode yourself.

Platform models work the same way: a loose name such as "qwen 3.8 flash next" is matched, whatever the case, spaces, dots, hyphens or word order.

Three smaller fixes belong to this change. With your own key the two verifiers now run on the same key as the run, not on the everyday model of the platform. The price estimate no longer reports unknown prices for runs on your own key, because ZeroLog bills nothing for them. And the Ultracode tools stay in the toolkit while Ultracode is on: before, loading three other tools could push them out, and the agent then called a tool it no longer had.

We checked this against a replica of the Anthropic model list and against the platform catalogue. We have not yet seen the id list of a real Anthropic key after this change, so whether a given Claude model appears depends on what your key lists at Anthropic.