bug: token cost seems inaccurate

[nikhilmaddirala]

I have some questions about this example conversation: https://opencode.ai/s/XPu4Slcg

1. It says Cost$0.25 Input Tokens 54541 Output Tokens 431 Reasoning Tokens — but according to openrouter the cost of z-ai/glm-4.5 is $0.20/M input tokens $0.80/M output tokens, which is much lower. Is this discrepancy due to a bug in opencode?
2. Where are the 50k+ input tokens coming from for my query and how can I inspect them?

Thanks in advance!

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [0.1.162] Openrouter price calculations likely does not consider cache read #534: OpenRouter price calculations likely does not consider cache read - specifically addresses discrepancies in OpenRouter cost calculations
• FEATURE REQUEST: Show tokens per file #1349: Shows tokens per file - requests ability to inspect where input tokens are coming from (your question 2)
• [Feature Request] Track Overall Token Usage (Usage Stats) #1634: Track overall token usage - related to understanding token consumption across sessions

Feel free to ignore if none of these address your specific case.

[rekram1-node]

Looking at openrouter I see:
Input: $0.6/M
Output: $2.20/M

see here

[nikhilmaddirala]

Looking at openrouter I see:
Input: $0.6/M
Output: $2.20/M

see here

But even w/ those numbers the cost seems incorrect...

When I click your link I see the same numbers I cited in my OP.

[rekram1-node]

@nikhilmaddirala I was looking at:

But yes you are correct and this appears to be a bug, thank you for reporting

[nikhilmaddirala]

[spoons-and-mirrors]

somewhat related to #1937 I believe

[rekram1-node]

just to provide insight into where the tokens are coming from, looks like vast majority are from the web fetch results, they can be rather expensive depending on website & content

[rekram1-node]

[automated] Closing due to 90+ days of inactivity. Feel free to reopen if you still need this!