dots3-note: 280B MoE agent learns memory via RL
Aiden_Tech_Ai · x · 2026-08-17
dots3-note preview is released as a step towards long-horizon agency. It features TEMPO, a new RL approach for agent training via self-critiquing.
Key Specs:
- 280B MoE with 16B active parameters
- 512K context window
- Multimodal (text, vision, audio)
It learns what to remember through RL and logs its own bugs. Open weights are available on Hugging Face, along with two new benchmarks: VibeSearchBench and VibeLifeBench.
More from Models
- Claude's coding 'one-shots' often just copy-pasted from humans — art_zucker · 2026-08-17
- Fable Coding Test: Elegant Solution Outperforming Opus 5 — arjunrajlab · 2026-08-17
- Claude's invisible text watermark uses SynthID, embedded token by token—and already beatable — APPSO · 2026-08-17
- Preview dots3-note: 280B open-weight multimodal model with 512K context — Aiden_Tech_Ai · 2026-08-17
- Why Qwen 3.8 27B Isn't Overthinking: Compared with GLM and DeepSeek — sukazu · 2026-08-17
- Meta Muse Glimmer 30B Native 512k Context: Architecture and Benchmarks — mr_il · 2026-08-17