ProviderStrategy: strict=True Only for OpenAI-Compatible Models

What it does

ProviderStrategy is an internal class in LangChain’s model-usage dispatch system that decides how to configure tool-calling per provider. The fix in 1.3.11 ensures strict=True is only applied to OpenAI-compatible models when using bind_tools(). Previously LangChain set strict=True for all providers, but only OpenAI and a handful of compatible providers (Azure OpenAI, Fireworks, OpenRouter) support it.

Why it matters

If you bound tools to a non-OpenAI model and saw malformed or missed tool calls, this was likely the culprit. bind_tools is one of the most common entry points in LangChain, so the blast radius was wide. Now the behavior is provider-aware and correct out of the box.

Example

$ Bind a weather tool to a Fireworks model and invoke it.

[
  {
    "name": "get_weather",
    "args": {"city": "Tokyo"},
    "id": "call_xxxxx"
  }
]

Common flags

bind_tools()

Binds tools to a chat model for tool-calling support

ProviderStrategy

Internal class that selects provider-specific tool config

strict=True

Manual override in the OpenAI API

create_openai_fn_runnable()

Creates a runnable calling an OpenAI function with structured output