Kaiming He's VISTA Achieves Perfect Score on ARC-AGI-3
GregKamradt · x · 2026-08-06
Kaiming He's team at MIT introduced VISTA, a visual harness that achieved a perfect 100% score on ARC-AGI-3, solving all 25 public games.
VISTA equips multimodal models with long-horizon visual reasoning capabilities. It allows the model to observe unfamiliar environments through high-dimensional sensory data like raw PNGs, recall past states via a lossless memory mechanism, and interactively explore the world to discover rules and achieve goals. Using Claude Opus 5.0 as the base model, the framework reached a 100% Relative Human Action Efficiency (RHAE).
More from Research
- SKILL-KD: Contrastive Skill Distillation for Weaker LLM Agents — ZhejiangUniversity · 2026-08-06
- Princeton Introduces Skill Entropy to Measure and Boost LLM Cross-Skill Reasoning — princetonu · 2026-08-06
- Tencent's WorldCycle: Self-Verifiable RL for Long-Horizon Video World Models — tencent · 2026-08-06
- NOLLI Benchmark: Diagnosing the English-Korean Performance Gap in LLMs — HAERAE-HUB · 2026-08-06
- Tencent Study: VLM Agents Face Severe Safety Risks from Stale Spatial Memory — tencent · 2026-08-06
- OneDayAgent: A Harness for Long-Horizon Autonomous Agents — zjunlp · 2026-08-06