rednote Open-Sources dots3-note: 280B-Parameter Multimodal Agent Model with 512K Context

On August 29, rednote's AI lab dots studio released the open-source multimodal agent model dots3-note Preview, targeting long-horizon, real-world agent tasks where requirements keep evolving. The model has 280 billion total parameters and 16 billion activated parameters, with a 512K context window, and is the first model in the dots3 family to jointly understand text, vision, and speech.

Confirmed

• Specs: 280B total / 16B activated parameters, 512K context length, open-source release
• Multimodal capabilities: joint understanding of text, vision, and audio
• Positioning: built for long-running workflows with evolving user needs that require continuous reasoning, execution, and dynamic adaptation; tackles the difficulty of maintaining plan coherence in long-horizon agent tasks
• Capability loop: pairs stronger coding, tool-calling, and end-to-end delivery capabilities to close the loop from task perception to action execution
• Official demo case: autonomously building a Xiaohongshu VR experience

Why It Matters

dots3-note is one of the few open-source models combining ultra-long context with joint understanding across text, vision, and speech, offering a new foundation model option for agent applications that need long-term memory and multimodal interaction. Its 512K context and sparse activation architecture (16B/280B) are purpose-designed for coherence in long-horizon tasks; follow-up community benchmarks will be worth watching.

2026-08-29 ~ 2026-08-30 · 5 related posts

Full story(2 episodes)→

• Episode 1: rednote Open-Sources dots3-note: 280B-Parameter Multimodal Agent Model with 512K Context(2026-08-29, 5 posts)
• Episode 2: dots3-note Demonstrated as Long-Horizon Agent in Wedding Planning Test(2026-08-30, 2 posts)

• rednote explores open-weight multimodal model with 512K context for long-horizon agents — aftahi_ai · 2026-08-29
• [source] dots3-note Preview: 280B Params, 512K Context for Long-Horizon Tasks — mhdfaran · 2026-08-29
• dots3-note Preview Launched: 280B Total/16B Active, 512K Context, Multimodal — SarahAnnabels · 2026-08-30
• dots3-note Preview: first dots3 model to jointly understand text, vision and speech — SarahAnnabels · 2026-08-30

1 near-duplicate retellings: SarahAnnabels