Daily Research Brief 2026-05-26

📊 Token usage: estimated from retrieval and writing scale.

Covers the latest AI research, open source and industry moves, updated daily.

Editor’s Note

The paper thread today: metaphorical video understanding (MetaphorVU), FlashAR post-training acceleration, LiteFrame long-video encoding, world-model-conditioned VLA (GigaBrain-0.5M), and benchmarks for trustworthy video LLMs.

1. Latest arXiv Papers

1. MetaphorVU: Towards Metaphorical Video Understanding — https://arxiv.org/abs/2605.25461

2. FlashAR: Efficient Post-Training Acceleration for Autoregressive Models — https://arxiv.org/abs/2605.09430

3. Audio-Visual Intelligence in Large Foundation Models: A Survey — https://arxiv.org/abs/2605.04045

4. LiteFrame: Lightweight Frame Encoding for Efficient Long-Video Understanding — https://arxiv.org/abs/2605.17260

5. MobileGym: A Verifiable and Highly Parallel Simulation Environment for Mobile Agents

6. From Model Scaling to System Scaling: Scaling the Harness for Long-Horizon Agents

7. Response-G1: Explicit Scene Graph Modeling for Proactive Embodied Response — https://arxiv.org/abs/2605.07575

8. GigaBrain-0.5M: World Model-Conditioned VLA for Robotic Manipulation — https://arxiv.org/abs/2602.12099

9. Trust-videoLLMs: A Comprehensive Benchmark for Evaluating Video LLMs — https://arxiv.org/abs/2506.12336

10. Artifact-Bench: Evaluating Multimodal Large Language Models on Artifact Understanding — https://arxiv.org/abs/2605.18984