Paper Overview
Research Area: Computer Vision Authors: Qiushi Han, Keya Hu, Linlu Qiu, Cathy Wu, Kaiming He Published: 2026-10-01 arXiv: 2610.02200
Abstract
We show that multimodal models possess strong reasoning abilities and that an appropriate harness can unlock their potential to solve tasks across diverse interactive environments. We introduce VISTA, a visual harness that gives a general-purpose multimodal model long-horizon vision. VISTA allows the model to directly perceive the environment through visual observations and maintains a lossless visual memory that preserves past observations in their original form. The model can actively retrieve these observations and reorganize its visual input as it reasons.
On ARC-AGI-3, VISTA improves Claude Opus 5.0's Relative Human Action Efficiency score from 40.68 to a perfect 100.00, with the model completing all 25 public games using 57.4% fewer actions than first-time human participants. VISTA's simple design allows it to extend to diverse visual environments with minimal adaptation. Across three additional benchmarks covering a variety of visual games and puzzles, it substantially outperforms baselines using the same underlying model with only a minimal harness. The results highlight VISTA's potential as a general visual harness for advancing multimodal agents in complex visual environments.
Key Results
- ARC-AGI-3: Relative Human Action Efficiency improves from 40.68 to 100.00 with Claude Opus 5.0
- Completes all 25 public games using 57.4% fewer actions than first-time human participants
- Outperforms same-model baselines on three additional visual game and puzzle benchmarks