Daily Research Brief 2026-08-10

📊 Token usage: estimated from retrieval and writing scale.

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

Editor’s Note

Today’s material states a judgment clearly: the breakthrough for long-horizon reliability is shifting from ‘swap in a stronger model’ to ‘move state out of context’. The Horizon Gap surveys 1,547 papers from 2024–2026, with the shared conclusion that outcome-level rewards fail quickly on long tasks; LongHorizon-Harness attacks the same problem from the harness side.

1. Latest arXiv Papers

1. The Horizon Gap: Planning, Memory, Execution, Training, and Evaluation for Long-Horizon LLM Agents — https://arxiv.org/abs/2608.06663

2. LongHorizon-Harness: Advancing Long-Horizon Agents for Real-World Tasks — https://arxiv.org/abs/2608.01964

3. DASH: Divergence-Adaptive Supervision Horizons for On-Policy Self-Distillation of Reasoning Models — https://arxiv.org/abs/2608.06243

4. Remember-R1: Process-Reward-Corrected Multimodal ‘Distributional Visual Forgetting’ — https://arxiv.org/abs/2608.01314

5. Vorch-Omni: Multi-Task Orchestration of Sight and Sound — https://arxiv.org/abs/2608.05803

6. Vorch-Streamer: Extending Human Audio-Visual Generation to Real-Time Long-Form Streaming — https://arxiv.org/abs/2608.05663

7. Understand Before Detect: Vision-Language Learning for Omni-Domain Infrared Small Target Detection — https://arxiv.org/abs/2608.07015

8. Token Communication for Multimodal Large Language Model — https://arxiv.org/abs/2608.07279