Every atomic assertion extracted from the underlying record, ranked by evidence strength.
LangChain Core improved JSON `get_format_instructions` using `Opik Agent Optimizer`.
LangChain Core improved LangSmith tracing by accepting integer temperature values in `_get_ls_params`.
LangChain Core added support for built-in tools in Groq message content.
LangChain Core added support for VertexAI standard content.
LangChain Core added a warning for fallback GPT-2 tokenizer usage.
LangChain Core improved error messages for missing titles in JSON schema functions.
LangChain Core added support for Python 3.14.
LangChain Core version 1.4.0 was released.
LangChain Core added content-block-centric streaming (v2).
LangChain Core's streaming v2 (stream_v2/astream_v2) is marked as beta.
LangChain Core added support for passing lists of ToolOutputMixin instances to _format_output.
LangChain Core deprecated `function_call` in favor of `tool_calls` for `get_buffer_string`.
LangChain Core added PEP 702 __deprecated__ attribute support to the @deprecated decorator.
LangChain Core automatically counts and stores metadata for tool call count.
LangChain Core's BaseTool now supports extras.
LangChain Core added chat model and LLM invocation parameters to traceable metadata.
LangChain Core added LangSmith integration metadata to `create_agent` and `init_chat_model`.
LangChain Core added `usage_metadata` to metadata in `LangChainTracer`.
LangChain Core started using `uuid7` for run IDs.
LangChain Core added ChatBaseten to its serializable mapping.
LangChain Core added support for imputing placeholder filenames for OpenAI file inputs.
LangChain Core added support for OpenAI tool search.
LangChain Core added a ChatAnthropicBedrock wrapper.
LangChain Core added the langchain-openrouter provider package.
LangChain Core added support for Google Maps grounding in the GenAI block translator.
LangChain Core minted langchain-model-profiles and added a profile property to BaseChatModel.
LangChain Core model profiles data is now distributed across packages.
LangChain Core includes original block type in server tool results for Google GenAI.
LangChain Core added ContextOverflowError which is raised in Anthropic and OpenAI.
LangChain Core model profiles added text_inputs and text_outputs.
LangChain Core added support for counting tokens from tool schemas approximately.
LangChain Core added support for scaling approximate token counts by reported usage.
LangChain Core added multimodal support to approximate token counting.
LangChain Core added 'approximate' alias in place of `count_tokens_approximately`.
LangChain Core added an XML format option for `get_buffer_string`.
LangChain Core added support for custom message separators in `get_buffer_string()`.
LangChain Core added proxy support for Mermaid PNG rendering.
LangChain Core added support for drawing subgraphs using `pygraphviz`.
LangChain Core added anti-SSRF hardening.
LangChain Core hardened `load()` against untrusted manifests.
LangChain Core restored cloud metadata IPs and link-local range in SSRF policy.
LangChain Core added more sanitization to templates.
LangChain Core hardened checks for txt files in deprecated prompt loading functions.
LangChain Core deprecated `prompt.save` and `load_prompt` methods.
LangChain Core removed `BaseMemory` (moved to `langchain-classic`).
LangChain Core removed items marked for removal in `schemas.py`.
LangChain Core removed `function_calling.py` utilities marked for removal.
LangChain Core removed `pydantic_v1/`.
LangChain Core removed `get_relevant_documents`.
LangChain Core removed globals from `langchain-v1` and updated globals in `langchain-classic` and `langchain-core`.
LangChain Core removed deprecated items.
LangChain Core improved performance by using reference counting for storing inherited run trees.
LangChain Core improved performance by reducing streaming metadata.
LangChain Core moved `BaseCrossEncoder` to `langchain-core`.
LangChain Core improved error messages for non-JSON-serializable tool schemas.
LangChain Core delegates to core implementation on invoke when streaming=True for OpenAI and Anthropic.
LangChain Core restores add_user_message and add_ai_message methods.
LangChain Core fixes PydanticOutputParser's get_format_instructions for v1 models.
LangChain Core overrides streaming callback if the streaming attribute is set.
LangChain Core fixes string content when streaming with output_version="v1".