Shanghai AI Lab's Atria Dawn: a research agent that flags what it can't verify

aliscodes · x · 2026-09-14

Shanghai AI Lab released Atria Dawn Preview, an agentic foundation model for long-horizon research tasks, with website, API, GitHub, Hugging Face and ModelScope availability.

The author tested it with a real task: find the 5 best-selling EVs of 2024 with prices and sources, build a comparison page, and flag anything unverifiable. It ran a dozen searches, pulled figures from real sources, built and opened the page — and explicitly flagged two prices it couldn't confirm instead of hallucinating them. The author sees this gap-flagging behavior as valuable for research work.

Related event: Shanghai AI Lab Releases Atria Dawn Preview, an Open-Source Research Agent Foundation Model(7 posts)→

Original post →

More from coding & agent

coding & agent channel →