# Symptom

A model-not-found style error from `init_chat_model("some-model")`, or the wrong provider gets picked.

## Confirm

Print the exact string you pass. Compare it character by character with the docs' model list.

## Cause

- Typo or invented variant (`gpt-5-5` vs `gpt-5.5`).
- Bare name ambiguity: inference picks a provider, and it may not be the one you meant.
- A proxy or gateway in front that restricts or renames models.

## Fix

Use the explicit form:

```python
model = init_chat_model("openai:gpt-5.5")
model = init_chat_model("anthropic:claude-sonnet-4-6")
```

Copy model identifiers from the docs, never from memory. If behind a proxy, check the proxy's allowed model names.

## Verify

Construct the model and run one minimal invoke. If it resolves, the string was the problem; pin the working string in config so it does not drift.