Tencent Hy3 Ranks 2nd in Open-Source Frontend Code Arena but Lags in Agent Arena
LMArena released its latest leaderboard data for the Frontend Code Arena and Agent Arena. Tencent's open-weight model Hy3 showed strong performance in frontend code generation, ranking 2nd among open-source models, but lagged in the Agent Arena based on over 8,000 real-world sessions, ranking 25th overall. This highlights a significant gap across different capability dimensions.
Confirmed
In the Frontend Code Arena, Hy3 ranks 2nd among open-weight models and 16th on the overall leaderboard. In specific vertical capabilities, the model entered the top 20 in reference-based design, game development, simulation, and content creation tools.
In the Agent Arena, Hy3 was tested through more than 8,000 live agentic sessions. It ranks 25th on the overall leaderboard and 5th among open-weight models. Furthermore, data reveals that the model's performance across various sub-categories shows a notably divided pattern.
Why it matters
The disparity in performance between static code generation and dynamic agentic tasks shows that strong frontend coding capabilities do not directly translate to overall agentic competence. This provides the community with a critical reference for evaluating and selecting open-source models.
2026-07-22 ~ 2026-07-22 · 8 related posts
Primary sources
- Tencent Hy3 ranks 25th in Agent Arena after 8,000+ live sessions — arena · 2026-07-22
- [source] Open-Weight Model Hy3 Ranks #16 in Frontend Code Arena — arena · 2026-07-22
- Hy3 ranks No. 2 among open-weight models in Frontend Code Arena — arena · 2026-07-22
- Tencent’s Hy3 ranks No. 5 among open models in Agent Arena — arena · 2026-07-22
- LMSYS Frontend Code Arena: Open-Weight Hy3 Ranks #2 — arena · 2026-07-22
- Tencent Hunyuan’s Hy3 ranks #25 in Agent Arena and #2 among open models in Frontend Code Arena — arena · 2026-07-22