The Qwen3.6b model in fireworks through initchatmodel reporting hugely inflated tokens

OSS Product Help LangChain
1

I am tracking token usage across all models. I have been using fireworks via the init_chat_model package.

Every model reports the correct token usage, except accounts/fireworks/models/qwen3p6-plus

I was able to fix it by setting model_kwargs={“reasoning_effort”: “none”},

it was working fine, but yesterday all of a sudden, it started reporting hugely bloated tokens again,

This is what i mean, actual tokens used 3800, repored tokens, 1229292929,

How do i fix this

2

Hi @JoAmps, would you please open an issue on the langchain repo with a MRE? That way we can investigate further. Thank you!