Sonnet-5.5 through `ChatOpenAI`

Hi LangChain Community,

A while ago Anthropic launched 5.5 and I was trying to access to access it thorugh ChatOpenAI wrapper. The migration docs explicitly state that on passing thinking.type as disabled we would be encountering a 400 error. However, when i pass it as disabled via the extra_params arg, I dont see the 400 error. I want to know validate if this paramter is actually being picked up or not. Is there any abstraction to hide this error and continue with the default params?

Thanks

@saqlain2204 why are you using ChatOpenAI versus ChatAnthropic?

We are routing our requests through litellm proxy

because the apis are openAI compatible, we are using OpenAI ChatModels

@saqlain2204
Why don’t you directly use ChatLiteLLM | langchain_litellm | LangChain Reference?