Tencent’s Hy3 ranks No. 5 among open models in Agent Arena
arena · x · 2026-07-22
Tencent’s Hy3 lands at #5 among open-weight models in Agent Arena and is also #2 among open models in Frontend Code Arena.
What the benchmark says
- Overall Agent Arena rank: #25 with -2.2% net improvement.
- Strength: tool use, especially recovering from CLI/bash errors.
- Weakness: steerability — it struggles to course-correct when users push back.
- The benchmark runs on millions of real-world long-horizon agent tasks, including web search, filesystem, terminal, slide decks, app building, and document analysis.
The result positions Hy3 as a strong open model for agentic workflows, but not a leader on overall user satisfaction or steerability.
More from coding & agent
- Harnesses widen what LMs can do, but may not improve compositional generalization — a1zhang · 2026-07-23
- A workflow for sharing agentic coding artifacts with teammates and pushing them to Git — rseroter · 2026-07-23
- An open-source /no-ai-slop skill removes 20+ common AI writing patterns — petergyang · 2026-07-23
- Open-source /no-ai-slop skill targets more than 20 common AI writing patterns — robleclerc · 2026-07-23
- LLMs are erasing framework-first identity labels among programmers — generativist · 2026-07-23
- Rork says Max App can build and preview iOS apps directly on iPhone — rudrank · 2026-07-23