Skip to content

He Kaiming's team proposes NAT-ARC, a pure-vision approach to solving ARC

2 reports 1 reporting sources first reported Updated 2026-10-08

Understand the event

Report summary

He Kaiming's team proposes NAT-ARC, a pure-vision ARC solving approach that does not rely on an LLM, using a MAE-pretrained encoder on ImageNet and then transferring it to abstract grid reasoning. The best-performing single model has 0.6B parameters and achieves 63.4% pass@2 on ARC-1, reaching 70.2% when ensembled, about one quarter of the parameter count of the specialized LLM approach The ARChitects (an 8B model, 71.6%).

From QbitAI

Read next

More changes in today’s briefingGemini Connects to Google Wallet for Personal Finance Insights

Report timeline

Follow the reports to explore different sides of the event.

10/8
  1. Kaiming He team proposes NAT-ARC, using ImageNet pretraining to solve the ARC challenge with vision models

    Kaiming He's team proposed NAT-ARC, a purely visual approach to solving ARC. By adding ImageNet MAE pretraining before the VARC pipeline, its best single model achieved 63.4% pass@2 on ARC-1, rising to 70.2% with an ensemble.

10/1
  1. Kaiming He's team proposes NAT-ARC: ImageNet pretraining for a purely visual ARC solver

    Kaiming He's team proposed NAT-ARC, a purely visual ARC-solving system that does not rely on LLMs. It pretrains an encoder with MAE on ImageNet and transfers it to abstract grid reasoning. The best single model has 0.6B parameters and reaches 63.4% pass@2 on ARC-1; an ensemble reaches 70.2%, with about one-quarter the parameters of the specialized LLM system The ARChitects, which uses an 8B model and reaches 71.6%.

Keep up with AI, one email a day

Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue