VSAS-Bench (Synchronous): leaderboard

VSAS-Bench (2026) tests vision-language models as streaming visual assistants on 92 videos with dense per-second annotations, under the synchronous protocol where the stream waits for the model. A GPT-5 judge scores each timestep's response against the reference; mean accuracy (%) over Present, Cumulative and Future tasks, higher is better.

Metric: Mean average accuracy (%) over all Present, Cumulative and Future tasks under the synchronous protocol (the model runs in lockstep with the stream, so inference latency does not count; video VLMs adapted with the sliding-window memory policy), over VSAS-Bench's 92 videos with dense 1 FPS free-form annotations; each timestep's response is judged against the annotation by GPT-5 (medium reasoning) on a 0-1 scale, shown as %; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5Low77.1#185
2Qwen 3 VL 32B Instruct59.5#607
3InternVL3-38B55.8#948
4Qwen 3 VL 8B Instruct54.3#1043
5Qwen 3 VL 4B Instruct51.6#1348
6Dispider46.8#3834
7InternVL3-8B46.3#1795
8Qwen 3 VL 2B Instruct42.4#2389
9InternVL3-2B36.9#3318
10StreamBridge [VSAS-Bench checkpoint unspecified]36.8
11Flash-VStream-7B13.5
12VideoLLM-Online11.3

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=vsas-bench-synchronous · How It Works · Data refreshed daily, snapshot 2026-11-22.