01
Foundation models
Build high-quality foundation models with stronger generalization and transfer.
More capable, efficient, and controllable models for real applications
We work across the full model lifecycle to improve capability, efficiency, and usability in real-world deployments.
01
Build high-quality foundation models with stronger generalization and transfer.
02
Design efficient, compact architectures that are practical to deploy.
03
Reduce latency while improving stability, safety, and throughput.
Build and process high-quality datasets
Efficient training and algorithm innovation
Align objectives and evaluate comprehensively
High-performance deployment and iteration
Architectures, strategies, and evaluation frameworks for practical small models.
2026-08-06
Optimization methods that improve throughput and reduce latency across settings.
2026-03-18
A common evaluation and data foundation for verifiable model development.
2025-06-26
Explore the next step in model research
Push the boundaries of capability, efficiency, and usability.
Explore model research