dots-studio has released the first open-weight model in its dots3 family, named dots3-note preview. This Mixture-of-Experts model features 280B total parameters with 16B activated and supports up to 512K tokens of context.

  • It is a multimodal model capable of understanding text, images, video, and audio while producing text outputs.
  • The architecture is optimized for general knowledge, mathematical reasoning, tool use, code generation, and long-context processing.
  • As the most lightweight member of the dots3 family, it balances capability with lower latency and inference costs compared to other potential family members.