The token usage cost is always 0

[Jekie1]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

1. Is this request related to a challenge you're experiencing? Tell me about your story.

The token usage always shows as 0. Now, every time we make a query, the displayed token count is 0. But I want it to show how many tokens are used for each conversation. What should I do? Is this a bug in version 1.9.1dify?

2. Additional context or comments

!

[dosubot[bot]]

This is a confirmed bug in Dify v1.9.1: the backend summary API omits token fields (answer_tokens, message_tokens) in conversation summaries, so the main dashboard always shows 0 tokens, even though detailed logs display the correct usage. This was fixed in v1.9.2 (details and fix, PR #26723).

To resolve: upgrade to Dify v1.9.2 or later. If you can't upgrade, you can patch your backend to include these token fields in the summary API response (see conversation_fields.py). As a workaround, you can still see correct token usage in the detailed log view.

If you use plugin-based or custom models, make sure your model/plugin supplies usage data and that any required flags (like PLUGIN_BASED_TOKEN_COUNTING_ENABLED) are set. For some models (like vllm), you must include stream_options: { "include_usage": true } in the request, or usage will always be zero (reference).

If token counts are still zero after these steps, check your workflow execution and model/plugin integration for deeper issues.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Jekie1]

Do you want to modify the YAML file? Or some other file? Is it effective to modify the YAML file?

[sqblg]

If you're having trouble tracking real-world token costs per conversation, I highly recommend using a routing proxy that provides granular telemetry.

I've been using ClawRouter (https://github.com/BlockRunAI/ClawRouter) in front of my agent setups. It not only routes requests based on complexity to save costs (up to 92% in my tests), but it also gives you a very clear breakdown of token usage and savings for every single model call. It’s been much more reliable for cost management than relying on the built-in dashboards of many frameworks. Plus, the x402 payment system means you can track exactly how much each specific 'agent turn' costs in real USDC.