[Question] Is it too inefficient and expensive to use?

[ttthree]

Questions

I just tried with open-router and claude-sonnet-4.5 model. A simple "deep-research" ran for a few minutes of browser agent to read websites, far away from completion and it already cost me $20, so i have to stop it ...

[Wendong-Fan]

hey @ttthree , for deep research tasks the agent would go through many websites to fetch information so it could consume a lot of token

[lightaime]

We are working on supporting prompt caching and tool call output offloading. C.c @Wendong-Fan which will cut do the cost significantly

[ttthree]

Thanks! Single call with 300K tokens input- this is $2 with Sonnet 4.5. I assume you put entire web page (or more) in the context, which is inefficient on token use, even prompt caching won't help here. Better approach would be extracting relevant snippets first.

[Wendong-Fan]

thanks @ttthree for the suggestion! We now provide the full DOM information to the LLM to ensure there’s no information loss that could lead to task failures, we’re planning to optimize tool-call token usage (see https://github.com/camel-ai/camel/issues/3675 ), and we expect to complete the implementation early this week. Once this optimization is integrated into Eigent, token consumption should be significantly reduced