VSAS-Bench: Real-Time Evaluation of Visual Streaming Assistant Models

TL;DR

VSAS-Bench introduces a comprehensive framework and benchmark for evaluating real-time visual streaming assistant models, emphasizing metrics like proactiveness and consistency in diverse, temporally dense video tasks.

Contribution

It provides a new benchmark with standardized evaluation protocols and extensive annotations, enabling large-scale assessment of streaming VLMs' capabilities and trade-offs.

Findings

Conventional VLMs can be adapted to streaming without extra training.

Adapted models outperform recent streaming VLMs, e.g., Qwen3-VL-4B beats Dispider by 3%.

The framework reveals key factors affecting accuracy-latency trade-offs.

Abstract

Streaming vision-language models (VLMs) continuously generate responses given an instruction prompt and an online stream of input frames. This is a core mechanism for real-time visual assistants. Existing VLM frameworks predominantly assess models in offline settings. In contrast, the performance of a streaming VLM depends on additional metrics beyond pure video understanding, including proactiveness, which reflects the timeliness of the model's responses, and consistency, which captures the robustness of its responses over time. To address this limitation, we propose VSAS-Bench, a new framework and benchmark for Visual Streaming Assistants. In contrast to prior benchmarks that primarily employ single-turn question answering on video inputs, VSAS-Bench features temporally dense annotations with over 18,000 annotations across diverse input domains and task types. We introduce standardized synchronous and asynchronous evaluation protocols, along with metrics that isolate and measure distinct capabilities of streaming VLMs. Using this framework, we conduct large-scale evaluations of recent video and streaming VLMs, analyzing the accuracy-latency trade-off under key design factors such as memory buffer length, memory access policy, and input resolution, yielding several practical insights. Finally, we show empirically that conventional VLMs can be adapted to streaming settings without additional training, and demonstrate that these adapted models outperform recent streaming VLMs. For example, Qwen3-VL-4B surpasses Dispider, the best streaming VLM on our benchmark, by 3% under the asynchronous protocol. The benchmark and code will be available at https://github.com/apple/ml-vsas-bench.