Agent Frameworks
LangChain, LlamaIndex, and Pydantic
- highInflated Token Usage Reporting for Fireworks Qwen3p6-Plus with reasoning_effort='none'
- high_to_protocol_usage drops input_token_details in v3 streaming path, causing cache token counts to always be 0
- high`create_agent` should support bounded forced retry after tool invocation validation errors
- highClearToolUsesEdit does not persist cleared state with checkpointer, causing eviction every turn
- highResponses API: parsed_arguments leaks from ParsedResponseFunctionToolCall into AIMessage content, breaking the next turn
