LANGCHAIN PATCHES: ANTHROPIC STREAMING, MISTRAL EMBEDDINGS RETRIES, CORE IMPORT MOVE
LangChain shipped small but meaningful updates across Core, Anthropic, and Mistral adapters that affect streaming, stability, and import paths. [langchain-anth...
LangChain shipped small but meaningful updates across Core, Anthropic, and Mistral adapters that affect streaming, stability, and import paths.
langchain-anthropic 1.3.5 adds eager_input_streaming, drops forced tool_choice when thinking is enabled, and tightens cache field handling. These changes make Claude streaming snappier and tool use behavior less surprising.
langchain-mistralai 1.1.2 improves embeddings exception retry logic and fixes a NameError in RunnableWithFallbacks, plus bumps urllib3. Model profile metadata also gets more consistent.
langchain-core 1.2.19 moves BaseCrossEncoder into core and bumps Tornado, which may change import paths and dependency constraints in some projects.
Anthropic eager_input_streaming and tool behavior tweaks can change latency and tool-calling semantics in production agents.
Import moves in core and retry fixes in Mistral can break assumptions or mask flaky failures in embeddings-heavy pipelines.
-
terminal
Run an end-to-end Claude streaming test with eager_input_streaming enabled, both with and without tools and thinking, and watch for timing or partial output regressions.
-
terminal
Hammer Mistral embeddings with parallel loads to verify new retry behavior (backoff, idempotency, error surfaces) and measure tail latency.
Legacy codebase integration strategies...
- 01.
Search and update imports for BaseCrossEncoder now under langchain-core; consider pinning versions before a broader upgrade.
- 02.
Re-evaluate dependency constraints around Tornado and urllib3 if you vendor or proxy requests; retest timeouts and proxy behavior.
Fresh architecture paradigms...
- 01.
Adopt Anthropic eager_input_streaming for faster perceived response in chat UIs and tool-enabled planners.
- 02.
Use updated model profiles to standardize config across environments, reducing drift between staging and prod.