DeepSeek
A family of low-cost, capable AI models. DeepSeek V4.1 Flash is cheap per token and built for tool calling, which suits an always-on agent on a budget.
#What it is
DeepSeek is an AI lab that publishes large language models, many of them with open weights, at prices well below most frontier models. The one I use is DeepSeek V4.1 Flash, reached through OpenRouter, which is designed for agent and tool-calling workloads.
#Why I use it
An agent like Hermes is token-hungry on the input side. Every turn it sends the model its system prompt, the full list of tool definitions, recent chat history and any files it opened. The output is usually small: a short reply and a tool call or two. So the input price drives the bill, and that's where DeepSeek is cheapest.
At the time of writing (September 2026), V4.1 Flash on OpenRouter costs around US$0.12 per million input tokens and US$0.48 per million output tokens. Prices move, so check the current figure before you commit.
#Benefits
- Very low cost per message, which is what makes a chatty household agent affordable.
- Good at tool calling, the part that matters most for an agent.
- Caching discounts on repeated input, and an agent repeats its tool definitions every single turn.
#Trade-offs and alternatives
- Data jurisdiction. Depending on which provider serves your request, your prompts may be processed outside your country. Keep secrets and personal data out of what the agent sees.
- Occasional tool-call fumbles on complex chains. If that happens to you, Gemini Flash models are very reliable at tool use, but cost several times more per token.
- The nice thing about going through OpenRouter is that none of this is a long-term commitment: switching model is one command.