dots3-note Preview: 16B Active Parameters Model for Long-Horizon Agency
rohanpaul_ai · x · 2026-08-24
rednote's dots studio released dots3-note Preview, an open-weight multimodal model designed for long-horizon agency in real life. It features 280B total parameters (16B active), a 512K context window, and supports text, vision, and speech. Key capabilities include Recursive Self-Critique via TEMPO, providing earlier learning signals in long tasks, and Test-Time Learning that allows the model to explore, hypothesize, and update its memory in novel environments like Slay the Spire 2.
More from Models
- Chinese Open Models Surpass US in Research Usage — xeophon · 2026-08-24
- Qwen3.8 27B at 11.7 tok/s on RTX 4070 Ti + Mac Air — zannix · 2026-08-24
- Teutonic-II 110B Open Source Model Released with Full Dataset — const_reborn · 2026-08-24
- NVIDIA's Catanzaro: Planning to Release Stronger Internal Teacher Models — Dan_Jeffries1 · 2026-08-24
- Gemini Desktop to Get Avatars; Gemini 4 Preparation Detected — testingcatalog · 2026-08-24
- ToMoE paper converts dense LLMs to MoE without fine-tuning — pmttyji · 2026-08-24