Breaking

Tencent's Hy4 Tops Arena's Chinese Model Rankings in Agent Evaluation

November 5, 2026 at 19:30 EST

Foundation Models
AI Agents

Tencent's Hunyuan team says its Hy4 model now ranks first among Chinese models on Agent Arena, the leaderboard that tracks how AI models perform on agentic tasks drawn from real users.

Tencent Hunyuan · Agent Arena leaderboard

Hy4 ranks first among Chinese models on Agent Arena

The ranking is built from more than 12,000 real-world, long-horizon agentic sessions submitted by users. Tencent's model is also the only Chinese lab in the overall top five.

Top spot among Chinese models

12,000+

Real-world agentic sessions in the ranking

Top 5

Only Chinese lab in the overall top five

+5.4%

"Net" score; baseline and method not explained

Hy4 reported net score: +5.4%

One block = 1 percentage point. The reference column is empty because the comparison baseline was not disclosed.

+5.4%

Hy4 net

Not stated

Baseline (unknown)

How a long-horizon agentic session is scored

1. Plan
Break a real user task into steps

→

2. Use tools
Act across many tool calls

→

3. Track progress
Keep state over a long session

→

4. Rank
Sessions feed the leaderboard

Why it matters

Agent tasks test planning and multistep work, not single answers.
User-driven sessions are often seen as closer to practical use than fixed test sets.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.