dots3-note: 280B MoE Model with TEMPO for Long-Horizon Agents

iamrobotbear · x · 2026-08-15

DotsStudioAI released dots3-note preview, a 280B MoE model with 16B active parameters and a 512K context window. It features TEMPO, a new RL approach for long-horizon agent training via self-critiquing and test-time scaling. The model is designed for reasoning, exploration, and memory updates, with open weights and new benchmarks (VibeSearchBench, VibeLifeBench) on Hugging Face.

Related event: Xiaohongshu Open Sources 280B Multimodal Model dots3-note(14 posts)→

Original post →

More from coding & agent

coding & agent channel →