Guide
Hermes Agent + DeepSeek: Use DeepSeek V4 with Hermes
DeepSeek V4 Pro and DeepSeek V4 Flash offer exceptional coding capability and strong general reasoning at some of the lowest token prices available — a compelling choice for cost-sensitive Hermes deployments.
What Is DeepSeek?
DeepSeek is a Chinese AI lab that released a series of open-weights models (DeepSeek V2, V3, V4) that repeatedly matched or exceeded frontier models from OpenAI and Anthropic at a fraction of the inference cost. The April 2026 launch of DeepSeek V4 Pro brought strong SWE-Bench coding scores alongside an input price around $0.14 per million tokens — roughly 20x cheaper than Claude Sonnet at similar quality on many coding tasks.
Hermes Agent supports DeepSeek through its built-in deepseek provider, which reads the DEEPSEEK_API_KEY env var and routes to the DeepSeek API at https://api.deepseek.com/v1. DeepSeek models are also available via OpenRouter.
DeepSeek V4 Models
| Model ID | Best For | Cost (Input) |
|---|---|---|
deepseek/deepseek-v4-pro | Coding, analysis, complex tasks — flagship quality | ~$0.14/M tokens |
deepseek/deepseek-v4-flash | Fast responses, lower latency, budget-conscious use | ~$0.04/M tokens |
DeepSeek V4 Pro is the recommended starting point. At $0.14/M input tokens, it runs roughly 20x more messages per dollar than Claude Sonnet 4.6 for tasks where both perform similarly. V4 Flash trades some quality for even lower latency and cost, making it suitable for high-volume or time-sensitive workloads.
Option 1: Hermes Agent on OpenClaw Launch (Easiest)
DeepSeek V4 Pro and Flash are available in the OpenClaw Launch model picker. No API key setup required.
- Go to openclawlaunch.com/hermes-hosting and start a Hermes deploy.
- Select DeepSeek V4 Pro or DeepSeek V4 Flash from the model dropdown.
- Connect your channel and click Deploy. Your DeepSeek-powered Hermes Agent is live in roughly 10 seconds.
Option 2: DeepSeek API Direct (Self-Hosted)
Hermes's built-in deepseek provider reads DEEPSEEK_API_KEY and talks to DeepSeek's API directly:
export DEEPSEEK_API_KEY=sk-...
hermes inference set deepseek
hermes model set deepseek-v4-pro
# config.yaml equivalent:
# inference:
# provider: deepseek
# model:
# default: deepseek-v4-proGet your API key from the DeepSeek Platform. The API base URL is https://api.deepseek.com/v1 and the endpoint is OpenAI-compatible.
Option 3: DeepSeek via OpenRouter (Self-Hosted)
If you already have an OpenRouter key, you can access DeepSeek V4 without a separate DeepSeek account:
export OPENROUTER_API_KEY=sk-or-...
hermes inference set openrouter
hermes model set deepseek/deepseek-v4-pro
# config.yaml equivalent:
# inference:
# provider: openrouter
# model:
# default: deepseek/deepseek-v4-proDeepSeek Direct vs OpenRouter
| Feature | DeepSeek Direct | OpenRouter |
|---|---|---|
| Latency | Lowest (direct) | Slightly higher |
| Price | DeepSeek native pricing | OpenRouter markup (~10%) |
| Access to other models | DeepSeek only | 200+ models, same key |
| Model switching | Requires separate keys for others | One key covers all |
Use DeepSeek direct if you only run DeepSeek models and want the absolute lowest cost. Use OpenRouter if you want the flexibility to switch between DeepSeek, Claude, GPT, and others via Hermes's /model command.
Switching to DeepSeek at Runtime
/model deepseek/deepseek-v4-pro
/model deepseek/deepseek-v4-flashWhat's Next?
- Hermes Agent + OpenRouter — Access DeepSeek and 200+ other models with one key
- Hermes Agent + GLM — Another cost-efficient frontier model option from Z.AI
- Hermes Agent + Claude — Compare DeepSeek to Claude for Hermes workloads
- Hermes Agent + MCP — Extend your DeepSeek-powered agent with MCP tool servers