Verdict: You can currently run Hermes Agent with zero API costs by switching your model to the newly released Tencent Hunyuan-3 (HY3). Available for free through the News Portal and OpenRouter until July 21, 2026, this 295B-parameter powerhouse provides a rare window to run complex agentic workflows—including coding and tool use—without burning your token budget.
Last verified: 2026-07-07 · Status: Free window active · Deadline: July 21, 2026 · Best for: Cost-free coding, research, and skill building.
The 60-Second Setup: How to Switch Hermes to HY3
Switching Hermes Agent to use the free Tencent Hunyuan-3 model is a non-technical process that takes less than a minute. You do not need to edit your config.yaml manually.
Step 1: Open the Model Selector
In your terminal, run the following command:
hermes model
Step 2: Select News Portal or OpenRouter
Use your arrow keys to navigate the menu:
- Select News Portal (which currently hosts the subsidised free HY3 preview).
- Choose Tencent HY3 (295B MoE) from the list.
- Press Enter to confirm.
Step 3: Verify the Connection
Start a fresh chat and ask: What model are you running?. Hermes should confirm it is connected to Tencent Hunyuan.
Why Tencent Hunyuan-3 (HY3)?
Tencent HY3 isn't just another "cheap" model; it is a flagship-class Mixture-of-Experts (MoE) architecture designed specifically for agents.
| Feature | Specification | Why it matters |
|---|---|---|
| Parameters | 295B Total / 21B Active | Frontier-level reasoning with low latency. |
| Context Window | 256,000 Tokens | High-capacity for long research papers and codebase analysis. |
| SWE-bench Score | 74.4% (Verified) | Capable of fixing real-world software bugs autonomously. |
| Cost | $0.00 (until July 21) | Unlimited testing for the next two weeks. |
Source: Tencent Hunyuan Technical Card & OpenRouter Stats
Advanced Workflow: The "Brain + Worker" Strategy
The real power of HY3 in 2026 is its role as a Worker Model. Because it is exceptionally cheap (or free right now), you can use it to delegate high-volume sub-tasks while keeping a more expensive model (like GPT-5.5 or Claude 4.7) as the primary "Brain."
- Primary Model: Handles high-level planning and decision making.
- HY3 Worker: Executes the "grind"—searching the web, reading files, and writing boilerplate code.
This hybrid approach allows you to implement the Token Minimizer Playbook and reduce your overall AI spend by up to 90% even after the free window ends.
Practical Test: Building Skills with slash learn
One of the best uses for this free window is teaching your Hermes Agent new skills. Using the /learn command, you can provide a link to a technical guide or documentation, and HY3 will synthesize it into a reusable .md skill file.
Try this:
- Type
/learn <URL to a documentation page> - Wait 30-60 seconds for HY3 to process the 256K context.
- Hermes will save the new skill to
~/.hermes/skills/, ready for future use—all for $0 in tokens.
What this means for you
The current "Free AI" era is defined by aggressive subsidies from providers trying to win the 2026 agent market. By switching to Tencent Hy3, you can bridge the gap between "toy" chatbots and professional-grade agentic operating systems without the financial risk.
Action Item: Switch your Hermes model today. If you are building a free AI coding setup, HY3 is currently your most powerful asset.
Related reading
FAQ
Q: What happens after July 21st? A: The free window on News Portal and OpenRouter is expected to close. The model will likely move to its standard (but still very low) pricing of ~$0.07 per million input tokens.
Q: Is HY3 as good as Claude 5 or GPT-5? A: Not quite. In our tests, it sits slightly below the current frontier flagships in creative prose, but it is remarkably competitive in logic and tool-calling.
Q: Does it work with the Hermes Gateway (Discord/Telegram)?
A: Yes. Once you update the model via the CLI (hermes model), the change applies to every messaging channel connected to that Hermes profile.
Discussion
0 comments