Research at :Harvey:
Our research teams are pushing forward the frontier of intelligence for professional knowledge work. Our goal is to make models and agents more capable by solving open research problems in synthetic and human data, post-training, agent harness optimization, verification, and more. We publish our findings regularly and build in the open alongside the open-source community.
Models
Post-training open-weight models for long-horizon problems, and making agents more reliable through harness optimization, reward modeling, and verification.
View all
Evaluation
Synthetic data generation and open-source benchmarks, built by domain experts to measure and improve models and agents.
View all
Field Notes
Observations and experiments from our research teams, covering agent behavior, interpretability, the impact of AI and society, and what we're learning from the field.
View all
