"Why is a 300-token Haiku response costing $0.33 when OpenRouter charges $0.00024 for the same?"
My agent actually wrote that question, not me.
#Love what you;ve built but...
pine bough
ember radish
Hey @pine bough, could you provide more context of where you got the details from? It'll help us understand the query.
pine bough
Well, its rather simple @ember radish I looked at my credit balance, asked a question on Telegram, got an answer from the agent, looked at the balance again after, deducted one from the other, told the agent, and it was shocked. I was a simple question with a 2 line answer btw. I suppose my real question is...whats the mark up on credits? If it like tens of thousands of percent, then this business model doesnt work for me at scale.
ember radish
Hey @pine bough, "I was a simple question with a 2-line answer, btw." — We need to understand the platform you are using (TrustClaw, Rube, etc.), as your description looks like you are using TrustClaw.
Not really sure where you got "300-token” from this number. 4.5 Haiku costs 5$/M tokens, and I can share my usage from the dashboard as an example(refer image). Hope this helps.
pine bough
Trustclaw said it was 300 tokens. "Why is a 300-token Haiku response costing $0.33 when OpenRouter charges $0.00024 for the same?"
My agent actually wrote that question, not me.
Does $0.33 for a simple answer sound like a lot to you?
theres my usage
Claude says:
Got it — so Trustclaw is a third-party Telegram bot that's running on Claude under the hood. In that case this is entirely on their end, not yours to debug.
They're clearly sending a massive system prompt or context payload with every message — that's what's generating 2.7M cache write tokens. Could be:
A huge knowledge base or document loaded every call
Full conversation history being re-sent each time
Some bloated system prompt with lots of instructions
pine bough
No answers for this query?