Submission checklist

• This is a bug, not a usage question.
• I added a clear and descriptive title that summarizes this issue.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).
• This is not related to the langchain-community package.
• I posted a self-contained, minimal, reproducible example. A maintainer can copy it and run it AS IS.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-openrouter
• langchain-perplexity
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Related Issues / PRs

No response

Reproduction Steps / Example Code (Python)

import os
from langchain.chat_models import init_chat_model

# Ensure FIREWORKS_API_KEY is present in environment
os.environ["FIREWORKS_API_KEY"] = "your_fireworks_api_key_here"

# Initialize using standard dynamic model initialization
llm = init_chat_model(
    model="accounts/fireworks/models/qwen3p6-plus",
    model_provider="fireworks",
    model_kwargs={
        "reasoning_effort": "none"  # Workaround that no longer prevents the issue
    }
)

# Execute query
response = llm.invoke("Explain vector databases in simple terms")

# Inspect response outputs
print("--- RESPONSE METADATA ---")
print(response.response_metadata)

print("\n--- USAGE METADATA ---")
if hasattr(response, "usage_metadata"):
    print(response.usage_metadata)
else:
    print("No usage_metadata found on response object.")

Error Message and Stack Trace (if applicable)

No direct python exception is thrown during invocation; instead, the returned response metadata contains corrupted integer values:

{ "token_usage": { "prompt_tokens": 125657, "completion_tokens": 1229292929, "total_tokens": 1229418586 } }

Description

We are tracking token usage and cost accounting across multiple Fireworks-hosted and external models using standard LangChain callbacks and response usage_metadata.

All other Fireworks models behave correctly. However, starting recently (without any changes to our local code), the reasoning model accounts/fireworks/models/qwen3p6-plus has started returning massively inflated, corrupted token counts.

Previously, we were able to work around token accounting bugs on this model by explicitly setting "reasoning_effort": "none" in the provider's model_kwargs. However, the workaround no longer mitigates the issue, and the API response payload now returns extremely large, invalid integer values for usage metrics.

What we are doing: Instantiating the accounts/fireworks/models/qwen3p6-plus model via the dynamic init_chat_model utility with the reasoning_effort set to "none".

What we expect to happen: The model should return reasonable, accurate token counts matching the actual prompt size and completion size (e.g., ~400 to 600 total tokens for a standard query of ~300 words).

What is currently happening: The model returns massively inflated numbers (e.g., total_tokens: 1229418586), breaking downstream cost tracking and database operations.

System Info

OS: macOS (ARM64) Python Version: 3.13 langchain: >=1.2.15 langchain-core: >=0.3.0 langchain-openai: >=1.2.1 langchain-fireworks: >=1.3.1