The Tencent HY3 model just dropped — a new open-source Chinese AI that's free to use, and I tested it the day it launched. Strong on some agentic benchmarks, genuinely useful for agent tasks, but not a frontier coder. Here's the honest rundown: what it is, how it performs, and how to use it for free.
📺 Watch: Tencent HY3 tested — benchmarks, agent tasks and the HY3 Coder demo
Want every new model tested and wired into one system? That's the Agent OS inside the AI Profit Boardroom. → Join AIPB
What Is The Tencent HY3 Model?
Tencent HY3 is a new open-source model from China, designed specifically for agent capabilities. Technically it's a 295-billion-parameter mixture-of-experts model with 21 billion active parameters, so it's built to run tools, browse and handle agent workflows rather than act as a pure frontier coder.
How To Use Tencent HY3 For Free
Right now you can use HY3 completely free in a few ways: it's on OpenRouter as a free API (until July 21), it's available through Nous Research with Hermes agent, and it works in Kilo Code. Just search "HY3" on OpenRouter, grab the API key, and plug it in. Local running via LM Studio isn't live yet, but given the older models were small (~8GB) it may arrive soon.
📺 Watch: These NEW Chinese AI Models are INSANE!
Tencent HY3: My Honest Test Results
I built landing pages and several games with it. The landing page was genuinely not bad. The games, though, mostly came out basic and glitchy — an open-world game that isn't quite playable, a driving game with very basic city blocks, weird lighting. Compared side-by-side with Fusion, GLM 5.2, Claude, Fable 5 or Opus 4.8, the quality and detail just aren't on the same level. That's expected for a free open-source model — but worth knowing before you rely on it.
Where HY3 Actually Shines: Agent Tasks
Here's the flip side. Plugged into Hermes, HY3 is genuinely good at agentic work — browsing ("open Google" just works), learning skills, reading pages fast, and running tools. It's quick, and it codes landing pages better than some paid models I've seen. Two standouts: the HY3 Coder workflow (prompts on the left, live previews on the right, all on the free API) is excellent, and giving HY3 access to Remotion let it create solid animated videos. So it's not a frontier coder — it's a fast, free agentic model.
📺 Watch: New Chinese AI Model Is INSANE!
Is The Tencent HY3 Model Worth It?
Honest verdict: HY3 benchmarks well on some agentic evals (it even beats GLM 5.2 on Claw Eval), but in hands-on testing it doesn't feel frontier. For building things day to day I'd still pick GLM 5.2 first, then Qwen 3.7. But as a free, fast model for agent tasks — browsing, running tools, quick coding, Remotion videos — HY3 is well worth testing, especially while the API is free.
Run HY3 (And Every New Model) In The Agent OS
I plug every new model like Tencent HY3 straight into the Agent Operating System inside the AI Profit Boardroom — with the HY3 Coder built in and an HY3 Hermes profile ready to go. Whichever model wins next, you swap it in without rebuilding anything. → Join AIPB.
Frequently Asked Questions
What is the Tencent HY3 model?
A free, open-source Chinese AI model designed for agent capabilities — a 295B-parameter mixture-of-experts model with 21B active parameters.
Is Tencent HY3 free?
Yes — free on OpenRouter (until July 21), via Nous Research with Hermes, and in Kilo Code.
Is HY3 better than GLM 5.2?
In hands-on tests, no — GLM 5.2 produced nicer, more detailed outputs. HY3 is better suited to fast agent tasks than frontier coding.
The Bottom Line
The Tencent HY3 model is a free, fast, open-source model that's strongest at agentic tasks — browsing, running tools, the HY3 Coder, and Remotion videos — rather than frontier-level building. Not the best open-source model out there (GLM 5.2 wins that), but genuinely worth testing while it's free.
