Shanghai AI Laboratory's Agents-A1 achieves top-tier benchmark results not by expanding parameters, but by extending task sequence length during training.
