Tencent Open-Sources HY-Embodied-0.5-X: Tops 6 of 10 Embodied AI Benchmarks

Tencent Robotics X Laboratory and Hunyuan team jointly released and open-sourced a multimodal large model for embodied task deployment—HY-Embodied-0.5-X.

The model achieved rankings of 6 first places and 3 second places across 10 benchmark evaluations covering planning, spatial reasoning, embodied QA, visual reference, and trajectory understanding.

Despite having only 2 billion parameters, the model surpasses larger models like Claude-4.0-Sonnet and GPT-5.4 in long-horizon operation tasks, ranking third overall behind TAIROS-Planning and Gemini3.0 Pro.

The model can generate complete planning for tasks and adjust during execution based on environmental feedback, implementing a ReAct loop with rapid re-planning capabilities.

Open Source Links:

• GitHub: https://github.com/Tencent-Hunyuan/HY-Embodied-0.5-X
• Hugging Face: https://huggingface.co/tencent/HY-Embodied-0.5-X