Tencent's Hunyuan team has made its Hy4 model available on Design Arena, a platform where users can access and compare AI models, and the company says the new version brings its first vision capabilities.
Tencent Hunyuan · New model release
Hy4 arrives on Design Arena with its first vision capabilities
Tencent says the new Hy4 model brings major gains in instruction following, long-context understanding, deep research and long-horizon tasks, and is now open for direct comparison on a public AI model platform.
How the gains were built
Phase 1
Expanded pre-training
Broad knowledge is learned from a larger training base.
→
Phase 2
Expanded post-training
The model is refined for specific behaviors.
→
Result
Hy4 capabilities
Stronger instruction following and long-horizon work.
What the Hy4 line now covers
Instruction following
Major improvement claimed
Long-context understanding
Major improvement claimed
Deep research
Major improvement claimed
Long-horizon tasks
Multi-step planning across many actions
New
Vision input
First vision support in the Hy4 line
Why it matters
Core needs for document analysis and research assistance
Vision widens the set of tasks the model can handle
Developers can test the claims directly on the platform
What remains unverified
No independent benchmark scores in the announcement
No pricing details or full modality list
Gains versus earlier Hunyuan models and rivals are untested
Open questions
Hands-on evaluation will decide how far Hy4 extends
Complex long-horizon tasks
Accuracy has not been detailed.
Real-world vision inputs
Performance has not been described.
Head-to-head standing
Needs
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...