VISTA: A Visual Harness for Reasoning in an Interactive World
cs.AI, cs.CV
Submitted: 2026-10-01
Updated: 2026-10-01
Comments: Tech report. An early version of this manuscript was in a blogpost published in Aug 5, 2026: https://vista-research.github.io/
Code: https://github.com/joshhhhhan/VISTA
Project page: https://schema-harness.github.io
License: http://creativecommons.org/licenses/by/4.0/
The gist: 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.
Terminology
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 also allows it to extend naturally to diverse visual environments with minimal adaptation. Across three additional benchmarks covering a diverse range of visual games and puzzles, it substantially outperforms baselines using the same underlying model with minimal harnesses. Our results highlight VISTA's potential as a general-purpose visual harness for advancing multimodal agents in complex visual environments.
Sources
- ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence
- On the Measure of Intelligence
- ARC-AGI-2: A New Challenge for Frontier AI Reasoning Systems
- ARC Prize 2025: Technical Report
- Screenshots or Tools? Eliciting Tool Use and Managing Multimodal Context in Hybrid GUI-MCP Computer-Use Agents
- GLM-5: from Vibe Coding to Agentic Engineering
- Less is More: Recursive Reasoning with Tiny Networks
- H-ARC: A Robust Estimate of Human Performance on the Abstraction and Reasoning Corpus Benchmark
- Tycho: Active Abstraction with Programmatic World Models for ARC-AGI-3
- What Memory Do GUI Agents Really Need? From Passive Records to Active Task-Driving States
- Qwen-CUA: Native Computer Use for (almost) Everything
- WebGPT: Browser-assisted question-answering with human feedback
- MemGPT: Towards LLMs as Operating Systems
- UI-TARS: Pioneering Automated GUI Interaction with Native Agents
- Do Coding Agents Need Executable World Models, Simplification, and Verification to Solve ARC-AGI-3?
- Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm
- Scaling Instructable Agents Across Many Simulated Worlds
- SIMA 2: A Generalist Embodied Agent for Virtual Worlds
- Thinking with Images for Multimodal Reasoning: Foundations, Methods, and Future Frontiers
- Hierarchical Reasoning Model
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems