SVCBench: leaderboard
BUAA CoLA Lab streaming video commentary benchmark: temporal reasoning, counting and narration over streaming video; GPA (Global Performance Average) %, human baseline 96.1.
Metric: GPA (Global Performance Average, %). Source: buaa-colalab.github.io. Saturation forecast: Around October 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 37 | #103 |
| 2 | Seed 1.8 | 36.2 | #143 |
| 3 | Qwen 3.5 35B A3B | 29.3 | #268 |
| 4 | GPT-5.4 | 29.1 | #83 |
| 5 | Kimi K2.5 | 26.4 | #152 |
| 6 | Qwen 2.5 VL 7B | 19.1 | #641 |
| 7 | Molmo2-8B | 8.5 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=svcbench · How It Works · Data refreshed daily, snapshot 2026-11-05.