Today for AI
HOT RADAR
llmHEAT 8.5°

AI CLUSTERED EVENT · 10/11/2026

He Kaiming Team Releases VISTA: Visual-Native Harness Framework Enables Claude Opus 5.0 to Score Perfectly on ARC-AGI-3

1 reports archived1 independent sourcesupdated 10/11/2026, 16:59:06
Synthesis & Latest Updates
1 Sources Cross-Validated

He Kaiming's team introduced VISTA, a visual-native harness framework that preserves raw visual information and allows models to revisit and zoom in on details during reasoning. This approach enhances existing multimodal models without retraining, enabling Claude Opus 5.0 to achieve a perfect score of 100 on the ARC-AGI-3 benchmark with 57.4% higher action efficiency than human baselines.

LATEST/He Kaiming's team introduced VISTA, a visual-native harness framework that preserves raw visual information and allows models to revisit and zoom in on details during reasoning. This approach enhances existing multimodal models without retraining, enabling Claude Opus 5.0 to achieve a perfect score of 100 on the ARC-AGI-3 benchmark with 57.4% higher action efficiency than human baselines.

HEAT TRENDHourly heat curve

9 fully observed hours
Now
8.5
Peak
10.423:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. 量子位官网T2·85 pts
    • VISTA is a visual-native harness framework that allows models to directly observe, save, and revisit raw visual frames during reasoning, rather than relying solely on abstracted text or code descriptions.
    • Without retraining, pairing VISTA with Claude Opus 5.0 achieved a perfect score of 100 across all 25 public games in ARC-AGI-3, while GPT-5.6 Sol scored 99.
    • The method significantly improved action efficiency, requiring 57.4% fewer game actions than the first-play human baseline, validating the effectiveness of visual memory in interactive reasoning.