Based on the published benchmark results, Hy3 appears to be in the same tier as models like DeepSeek v4 and GLM-5.1.
Beyond the benchmarks, Tencent also released results from 312 real-world workflow tasks.
Hy3 scored 2.67/4 versus 2.51/4 for GLM-5.1, with the biggest reported improvements in frontend development, CI/CD, and data/storage.
I'm treating those results as a useful signal rather than proof until more independent testing comes out. What I'm more interested about is whether those gains hold up once people start throwing real projects at it.
It's already available on OpenRouter, so we should start seeing more hands-on feedback before long.
Weights: https://huggingface.co/tencent/Hy3