Tencent's Hy4 model has debuted at No. 14 overall on the Agent Arena leaderboard, a ranking built from real-world agentic tasks, and the model could tie for first place among open models if its weights are published.
Agent Arena leaderboard · Tencent Hunyuan
Tencent's Hy4 debuts at No. 14 overall on agentic tasks
Hy4 ranks 14th on a leaderboard built from long-horizon, multi-step agent sessions. The model is proprietary today, but Agent Arena says it would tie for first among open models if its weights are released.
14th
Current overall rank for Hy4
8,500+
Long-horizon agentic sessions behind the ranking
Tie 1st
Among open models, only if weights are published
Scale of the evidence: 8,500+ sessions
Each block represents 500 sessions (17 blocks ≈ 8,500).
Sessions test whether a model can complete multi-step tasks over extended interactions, rather than answering single prompts.
Positive signal
Places Hy4 among the stronger entries on the platform.
Results reflect real-world agentic usage rather than single prompts.
Projected to tie for first among open models once released.
Open questions
Scoring methodology, task categories and confidence intervals were not disclosed.
Independent verification of the ranking has not been reported.
No release date or licensing terms for the weights.
Bottom line
The open-model standing depends entirely on a release that has not been scheduled.
Current status: proprietary, weights not publicly available.
Projected status: tied with Kimi K3.1 Pro for first among open models, if weights are published.
Until independent checks and an official Tencent release plan appear, treat the figures as the platform's own assessment of current performance.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...