Skip to content
Hot eventLive

何恺明团队NAT-ARC：纯视觉方案逼近ARC专用LLM

1 reports1 sources7 hr ago updated

Get the story

AI overview

2026-10-01，何恺明团队提出NAT-ARC，在VARC视觉流程前插入ImageNet MAE预训练，让模型从真实图像迁移到抽象格子推理。最佳单模型0.6B在ARC-1上pass@2达63.4%，集成后70.2%，接近8B专用LLM方案的71.6%。预训练打通了视觉路线的scaling瓶颈，模型越大效果越好。

Generated from reports · updated 5 hr ago

LatestOct 1
NAT-ARC纯视觉方案在ARC-1上集成达70.2%，逼近8B专用LLM的71.6%。

Timeline

Follow the coverage from different angles.

Oct 1, 2026
  1. 量子位
    何恺明团队NAT-ARC：纯视觉方案逼近ARC专用LLM水平

    何恺明团队提出NAT-ARC，在VARC视觉流程前插入ImageNet MAE预训练，让模型从真实世界图像迁移到抽象格子推理。最佳单模型0.6B在ARC-1上pass@2达63.4%，集成后70.2%，接近8B专用LLM方案的71.6%。预训练打通了视觉路线的scaling瓶颈，模型越大效果越好。

Heat trend

Current heat 8·Comparable peak 10（Oct 2）·Comparable change over 24 hours –

02.557.510Oct2Oct2Oct2Oct2

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.