I’m not going to lie, I am addicted to my AI agents. I am dreading being away from them in Europe but I know I can access them remotely, it’s just not the same as being nearby my agents. But my addiction is coming at a cost that I don’t like so I’m researching cost structures for my AI agents.
Here are a few websites you should become familiar with if you decide to embark on an agentic journey.
OpenRouter.AI
Open Router is THE place to go for your AI agentic needs. If you plan on testing/using different AI models then you will need to use OpenRouter because other ways will become very painful very quickly. But one of the things you will find on OpenRouter is the link below which shows you what LLMs are being used by people using OpenRouter.
https://openrouter.ai/apps/hermes-agent
This snapshot is from June 24th and will change by the time this post goes live.
The trends in the graph are interesting. The lime green (chart above) is Owl Alpha which is currently free. The blue is DeepSeek Flash and the gold is DeepSeek Pro. I started with Claude and GPT but now I’m using DeepSeek primarily and testing GLM 5.2.
The graph below shows my experimentation and transition from using GPT (light pink) to Claude (puke green) to DeepSeek (cyan) and notice how the cost rolls down even though I am doing more with DeepSeek.
The list below shows the most popular LLMs on OpenRouter. DeepSeek v4 Flash is the second most used LLM and I joined the wisdom of the crowd for cost reasons.
BenchLM.AI
There are many websites now tracking the costs of LLMs but I like https://benchlm.ai/llm-pricing because its fairly easy to read. The important thing to pay attention to is output cost.
The wisdom of the crowd in the chart above shows that most Hermes Agent users are using DeepSeek, GLM or Kimi which I believe are all low cost China LLMs. There is also usage from other LLMs but I suspect that China’s LLMs will rise to the top over time simply because of the cost/value proposition.
I am trying to keep my AI usage to under $100/month for 3 agents but I plan on adding more and the ultimate solution may be to buy my own hardware and run LLMs locally.
Share The Wealth
Are you addicted to AI agents yet?