Atria Dawn Preview: student-heavy team launches research-focused agentic base model
xiaohu · x · 2026-09-15
X user xiaohu spotlights Atria Dawn Preview, a base model aimed at research and complex agent tasks (paper on Hugging Face). Its claimed key feature is a verifiable experience pipeline: only experiences validated in real environments with metric checks and evidence loops get distilled into the model, including learning from failures. It reportedly predicts global weather a week ahead in 1 minute offline (unverified). Of 96 core members, 65 are students, formed by Shanghai AI Lab with Fudan, CAS institutes, Renmin University, East China Normal University and Harbin Institute of Technology. The model ships with a bound agent harness.
More from Models
- Researcher rebuts 'models know we're studying them' claim: they're passive computation — vishalmisra · 2026-09-15
- Swapping in Exa search makes DeepSeek V4.1 Flash consistently flagship-grade, tester finds — bookwormengr · 2026-09-15
- AI Overview auto-applies the 'creative writing' jailbreak, researcher shows — conitzer · 2026-09-15
- Solar Pro 4 stays free on Nous Portal until Sept 25 in Upstage partnership — keunwoochoi · 2026-09-15
- Bug Hunt Bench: multiple runs boost small-model bug detection but move frontier models just 1-2 points — PawelHuryn · 2026-09-15
- ZDTaichu5.0-9B, a 9B vision-language model with spatial reasoning, trends on Hugging Face — TaichuAI · 2026-09-15