benchmarks evals
The Fast Event Fell Between the Frames
FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.
Summary
FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.
FastBench contains 306 verified questions across eight domains and six capabilities, filtering out questions answerable at two frames per second and using three rounds of human inspection. Its strongest tested model, Gemini 3.5 Flash, scored 50.7%. Raising Qwen3-VL-8B sampling from two to twenty-four frames per second improved its score from 32.9% to 44.6%, but gains saturated as older history was compressed. The training-free ProactiveFrame baseline adjusts incoming rates and keeps recent dense observations alongside sparse history. It improved on sparse uniform sampling by 5.4 and 1.5 percentage points in the reported comparisons while remaining below oracle-guided focusing. Choosing when to look more closely remains part of the problem.
Why it matters
FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.
Limits and context
No additional limitation was separately recorded.
Key claims
FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.
Evidence: source-2026-10-09-005
Sources
- arXiv preprint 2610.12427, original v1arXiv / submitting research authors · primary research
Corrections
No corrections have been recorded for this story.