# GeekPark tests Tencent's Hy4 model on long Agent tasks and visual code

> Original title: 实测 HY4：腾讯最「夯」大模型，离一线还有多远？

- Source: GeekPark
- Published: 2026-11-09T03:57:03.000Z
- Bad Influence: https://badinfluence.news/items/kvgb52mpzlrwt5ozzcu67dgm0
- Original: http://www.geekpark.net/news/373644

## Summary

GeekPark tested Tencent's Hy4 model, released November 4, on long-horizon Agent tasks, visual understanding, and visual code generation. The report says Hy4 ranks in the first tier of open-source models but still trails leading closed models on high-difficulty reasoning and engineering tasks.