[Release] langchain-ai/langchain langchain-fireworks==1.2.0: langchain-fireworks==1.2.0

langchain-fireworks version 1.2.0 introduces several improvements including proper handling of `max_retries`, population of `usage_metadata` on streaming responses, and enhanced model profile data. The release includes dependency updates (langsmith, pillow, pytest), security patches (pygments CVE-2026-4539), and standardization of integration tests across the partners ecosystem. Key fixes address tool binding with strict mode, reasoning content handling, and model naming schema consistency.

Key Points

• •Honor `max_retries` parameter in Fireworks integration for proper retry behavior
• •Populate `usage_metadata` on streaming responses to provide token usage information during streaming
• •Upgrade langsmith from 0.4.31 to 0.7.31 for improved compatibility and features
• •Apply security patch for pygments (CVE-2026-4539) across all packages with version >=2.20.0
• •Standardize integration test invocation across all partner integrations for consistency
• •Add `text_inputs` and `text_outputs` fields to model profiles for better capability tracking
• •Implement `bind_tools(strict: bool)` parameter and fix `reasoning_content` handling
• •Refresh model profile data multiple times to ensure accurate model metadata
• •Bump langchain-core minimum version to 1.2.21 for required functionality
• •Fix model naming schema and test model names for consistency across the integration

Workflow Diagram

Concepts