Search papers, labs, and topics across Lattice.
The Ohio State University
2
0
6
Training on D3-Gym, a new dataset of real-world scientific environments, boosts Qwen3-32B performance on ScienceAgentBench by 7.8 points, rivaling proprietary models.
Multimodal agents can now reason, plan, and execute actions more effectively by integrating perception as a core component, not just an auxiliary interface.
