VSAS-Bench (Synchronous) - Future: leaderboard

VSAS-Bench (2026) Future tasks, which ask the model to anticipate upcoming events from ongoing visual cues in a streamed video, under the synchronous protocol; per-timestep responses judged against dense reference annotations, accuracy (%), higher is better.

Metric: Mean average accuracy (%) on Future tasks (predicting upcoming events from current cues) 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-5High85.1#178
2Dispider66.6#3815
3Qwen 3 VL 8B Instruct35.5#1029
4InternVL3-38B34.7#939
5Qwen 3 VL 32B Instruct31.4#606
6Qwen 3 VL 4B Instruct19.5#1332
7InternVL3-8B14.6#1782
8StreamBridge [VSAS-Bench checkpoint unspecified]14.1
9Qwen 3 VL 2B Instruct13.1#2367
10InternVL3-2B13#3290
11Flash-VStream-7B8.7
12VideoLLM-Online2.3

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=vsas-bench-synchronous-future · How It Works · Data refreshed daily, snapshot 2026-11-21.