- Who it helps
- Researchers evaluating multi-turn video assistants.
- Availability
- Public paper submitted October 29, revised November 3. Project links code and data; dataset terms restrict annotations to noncommercial academic research and evaluation, excluding training.
Connected questions
The benchmark has 326 videos and 2,476 turns. Some questions depend on a previously described visual episode; others depend on an earlier answer. The authors check that later questions need those earlier turns.
What the result establishes
The authors report that their strongest tested model scores 45.7% overall. This identifies difficulty within their evaluation; it does not rank all video use cases or establish production reliability.
VideoMemBench tests memory across video questions ↗VideoMemBench project ↗VideoMemBench evaluation repository ↗VideoMemBench dataset terms ↗
What you can do
- Read the linked primary evidence and its access limits before planning a trial. Open reference ↗
Where the evidence stops
Paper, project and artifact pages checked November 5. No evaluation executed; original video rights are separate from annotation terms.
Sources read. Sources read November 5, 2026 UTC. No third-party project, model inference or complete product workflow was executed.
Sources & further reading
Paper · paper
VideoMemBench tests memory across video questions ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
VideoMemBench authors · page checked 2026-11-05
GitHub · implementation
VideoMemBench evaluation repository ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
Oliiveralien · page checked 2026-11-05
Web · official
VideoMemBench project ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
VideoMemBench authors · page checked 2026-11-05
More sources reviewed (1)
Hugging Face · official
VideoMemBench dataset terms ↗
Primary page read November 5. Claims were checked against the page; the workflow was not executed.
VideoMemBench authors · page checked 2026-11-05
Coverage note: Original YouTube demonstrations and independent cross-platform attention were not established. No end-to-end workflow was executed.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
VideoMemBench tests whether video assistants can reuse earlier scenes and their own earlier answers.