Tencent Hunyuan Releases Hy3 Model With 295B Parameters and Agent Focus

Tencent Hunyuan released the Hy3 model.

The system uses a 295 billion parameter mixture of experts design.

Only 21 billion parameters activate during inference.

This efficiency matches much larger models.

Hy3 aims at agent workflows rather than general chat.

It already runs inside WeChat for over one billion users.

Release Details Meet Real Workloads

The company tested Hy3 on more than fifty internal business tasks.

Task success rose from 72 percent to 90 percent.

Average task time fell by 34 percent.

The model now handles code writing, office automation, and multi step planning.

Pure vision tasks remain a clear weakness.

Internal teams used the preview version for months.

Results guided changes before the public step.

Agent Focus Creates New Pressure Points

Most large models still optimize for broad conversation.

Hy3 instead targets repeated tool use and self correction.

It can review its own output and flag missing data.

This trait matters for agent loops that run without constant human review.

Rivals that focus only on raw scale now face a different benchmark.

WeChat integration gives Tencent immediate volume no other lab matches.

Efficiency Numbers Stand Out

The 21 billion active parameters deliver strong coding scores.

They also drive solid performance on office documents and slide generation.

A recorded demo showed the model create a full ten page presentation.

Another demo built a working HTML agent interface.

These examples rely on planning rather than image understanding.

The gap in vision keeps Hy3 from competing in multimodal benchmarks.

Limits Remain Clear

Vision shortfalls will limit some agent use cases.

Complex visual planning still needs human help or other models.

The company has not published independent third party scores.

All current numbers come from internal tests.

Users will watch whether public benchmarks confirm the same gains.

Next Signals To Track

Watch for external coding benchmarks in the next quarter.

Observe whether WeChat agent features expand beyond the current demo set.

Track competitor responses in the agent tooling space.

Each of these moves will show if the efficiency claim holds at scale.