langchain langchain-anthropic==1.7.11
- Source
- github.com
- Author
- github-actions[bot]
- Date

Context
langchain-anthropic 1.7.11, the package that connects LangChain applications to Claude, is the version to install before pointing such an app at Anthropic's Claude Fable 5.2 or Claude Mythos 5.2. The release notes list support for both models, and comparing the published package with version 1.7.10 shows what that support changes in practice.
The main change is , where an app asks the model to answer in a fixed data format. By default LangChain does this by forcing the model to call a tool whose inputs match that format. Version 1.7.11 adds Fable 5.2 and Mythos 5.2 to the package's list of models that do not accept a forced tool call, next to Fable 5.1, Opus 5.5 and Sonnet 5.5, so the package now asks for the tool without forcing it and checks the reply instead. In 1.7.10 neither new model was on that list. The release also separates refusals from formatting failures: when the model declines and returns no tool call, the app now gets a ModelRefusalError rather than a generic parsing error, so it can tell a safety refusal from a malformed answer.
Mythos 5.2 also now gets the request handling the package already applied to Fable models: it rejects non-default temperature, top_p and top_k settings with a clear error instead of sending them. The bundled model profiles record a 1 million input limit, a 128,000 token output limit and effort levels from low to max for both models. A smaller fix emits a chunk as soon as a thinking block starts, which a code comment ties to responses where the reasoning text itself is hidden, so an app can still show that the model is working. Anthropic's Fable 5.2 prompting guide covers the refusal stop reason that the new error type reports.
- structured output — Forcing a model's response to match a schema, so downstream code can parse it instead of guessing at prose.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
Checking sign-in…
Loading comments…






