Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent
Published:
Authors: Lei Bai, Zongsheng Cao, Yang Chen, et al., including Zelin Tan.
This work introduces Agents-A1, a 35B mixture-of-experts agent model trained by scaling long-horizon trajectories and heterogeneous agent abilities. The resulting model is competitive with much larger models across a broad set of long-horizon agent benchmarks.