What langchain-ai/langchain shipped
Written by FoxPlug from the repository's own releases; not affiliated with LangChain. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/langchain-ai/langchain; LangChain did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of October 5, 2026
What shipped
- Model profiles now declare mid-conversation system message support, letting callers make better decisions about when to keep system messages in place during multi-turn conversations. Pull request #41180
- With_retry now accepts a callable to define custom retry logic based on exception properties. Pull request #41158
- ChatFireworks now preserves reasoning tokens during streaming and tool loops, fixing losses in multi-turn interactions. Pull request #41093
- SummarizationMiddleware retries with conservative trimming when summarization overflows context limits. Pull request #41159
- Fixed annotation handling for Python 3.14 compatibility where annotations are evaluated lazily. Pull request #41059
- IPv6 SSRF security checks hardened in core. Pull request #41151
- HuggingFaceEndpoint now hides API tokens from repr output to prevent accidental credential exposure in logs and traces. Pull request #41047
- Hugging Face streaming test updated to use currently supported Scaleway model. Pull request #41095
- FastMCP minimum version bumped to 4.0.11 for MCP support. Pull request #41054
Why it matters
This week brings improvements to mid-conversation handling in multi-turn workflows, better control over retry behavior with custom logic, and critical fixes for streaming reasoning preservation and security. Several patch releases stabilize edge cases in context overflow, test reliability, and Python 3.14 support.
Changelog entry
- Langchain 1.4.4 released with SummarizationMiddleware context overflow retry. Release
- Langchain-core 1.6.9 released with callable support in
with_retry. Release - Langchain-core 1.6.8 released with IPv6 SSRF hardening. Release
- Langchain-fireworks 1.7.1 released with streaming and tool loop reasoning preservation. Release
- Langchain-huggingface 1.2.3 released with updated Scaleway model for streaming tests. Release
- Langchain-core 1.6.7 released with Python 3.14 hardening and Bedrock Converse fixes. Release
- Model profiles now declare mid-conversation system message support via declarative model name matching. Pull request #41180
- HuggingFaceEndpoint hides API tokens from repr output to prevent credential exposure. Pull request #41047
- With_retry accepts callables for exception-based retry decision logic. Pull request #41158
- ChatFireworks preserves and replays reasoning tokens in streaming and tool loops. Pull request #41093
- SummarizationMiddleware retries with conservative trimming on context overflow. Pull request #41159
Week of September 28, 2026
What shipped
- Langchain-text-splitters 1.1.3 released with dependency updates. Release
- Langchain-core 1.6.6 released with compatibility fixes and documentation corrections. Release
- Langchain-fireworks 1.7.0 released with prompt caching middleware support. Release
- Langchain 1.4.3 released with cache sanitization and Bedrock Mantle support. Release
- FireworksPromptCachingMiddleware added to improve prompt-cache reuse across agent turns. Pull request #38823
- ChatOllama reasoning parameter now requires ollama floor 0.6.3 for thinking levels support. Pull request #40986
Changelog entry
- Langchain-text-splitters: 1.1.3 with dependency updates Release
- Langchain-core: 1.6.6 with documentation fixes Release
- Langchain-fireworks: 1.7.0 with prompt caching middleware and timeout classification Release
- Langchain: 1.4.3 with cache sanitization for fallback models and Bedrock Mantle support Release
- Fireworks: prompt caching middleware for improved cache reuse in agent threads Pull request #38823
- Ollama: raised floor to 0.6.3 for ChatOllama reasoning parameter support Pull request #40986
- Langchain: sanitized cache settings for fallback models to prevent validation errors Pull request #40886
- Fireworks: mid-stream read timeouts now classified as retryable ModelTimeoutError Pull request #40874
Week of September 21, 2026
What shipped
- AWS Bedrock Mantle chat models are now accessible through
init_chat_model. Pull request #40837 - Create_agent now repairs invalid tool calls with error ToolMessage entries to prevent provider replay failures. Pull request #40530
- AzureChatOpenAI users on AKS can now automatically use workload identity without manual credential injection. Pull request #40532
- ChatFireworks now replays historical tool calls with malformed arguments as diagnostic JSON instead of forwarding invalid strings. Pull request #40818
- Long tool IDs in XML-formatted conversation strings are now abbreviated to first 32 characters to reduce token usage. Pull request #40792
- Documentation in AGENTS.md and package docs have been corrected for accuracy with repository structure. Pull request #40791
Changelog entry
- Bedrock Mantle chat models now supported in
init_chat_modelPull request #40837 - Create_agent repairs invalid tool calls with error ToolMessage entries Pull request #40530
- AzureChatOpenAI supports workload identity on AKS without manual configuration Pull request #40532
- Fireworks: malformed tool arguments preserved as diagnostic JSON Pull request #40818
- Long tool IDs abbreviated in XML conversation strings Pull request #40792
- Documentation accuracy improvements in AGENTS.md and package docs Pull request #40791
- Langchain-fireworks released 1.6.3 Release
- Langchain-core released 1.6.5 Release
- Langchain-openrouter released 0.2.9 Release