Research
Exploring the frontiers of artificial intelligence through five core research directions
Interaction as Intelligence
Advancing Human-AI Collaboration
We believe that the future of AI lies not in replacing humans, but in creating intelligent systems that enhance human capabilities through natural, intuitive interaction.
Model Architecture
Beyond Transformers
Exploring innovative neural architectures that can overcome the limitations of current transformer-based models and unlock new capabilities.
Foundation Model
Next-Generation Foundation Models
Building foundation models that demonstrate enhanced reasoning, understanding, and generalization across diverse domains and tasks.
MegaScience
Pushing the frontiers of post-training datasets for science reasoning.
OctoThinker
Exploring how different early pre(mid)-training strategies could bring impact to post-training stages, especially during the period of Reinforcement Learning (RL).
Thinking with Generated Images
A single LMM (Large Multimodal Model) that spontaneously generates and reasons with intermediate visual thoughts via a native long-multimodal thought process.
Agents
Autonomous AI Systems
Creating autonomous agents with sophisticated agency that can understand goals, plan actions, and execute complex tasks in dynamic environments.
Benchmark
Evaluation & Measurement
Developing comprehensive evaluation frameworks and benchmarks to accurately measure AI capabilities and drive continuous improvement.
