Xiaohongshu's (RedNote's) in-house AI lab, launched in early 2025 out of its LLM team and formerly known as Hi Lab.

First open-weight release was dots.llm1 (June 2025), a 142B-total / 14B-active MoE; dots.ocr (July 2025) is a 1.7B multilingual document parser.

Elevated to a first-level 'Dots' AI department inside Xiaohongshu in April 2026.

A dots-note-3.0 branch inside a bespoke proof harness posted a perfect 42/42 at IMO 2026; the published dots3-note Preview checkpoint is not that system.

Dots3-Note Preview (14 August 2026, Apache 2.0): 280B total / 16B active MoE with vision and audio encoders, 512K context, first open-weight dots3 model.

Xiaohongshu's (RedNote's) in-house AI lab, launched in early 2025 out of its LLM team and formerly known as Hi Lab.

First open-weight release was dots.llm1 (June 2025), a 142B-total / 14B-active MoE; dots.ocr (July 2025) is a 1.7B multilingual document parser.

Elevated to a first-level 'Dots' AI department inside Xiaohongshu in April 2026.

A dots-note-3.0 branch inside a bespoke proof harness posted a perfect 42/42 at IMO 2026; the published dots3-note Preview checkpoint is not that system.

Dots3-Note Preview (14 August 2026, Apache 2.0): 280B total / 16B active MoE with vision and audio encoders, 512K context, first open-weight dots3 model.