Understand this story
On November 11, 2026, Ant inclusionAI open-sourced ZoomGain on GitHub and simultaneously released the ZoomGain-4B and ZoomGain-9B model checkpoints. The method uses visual details that only become clear after magnification to generate questions about the entire image, constructing supervision signals for vision-language models. The same release also includes an evaluation benchmark of 1,000 questions covering 941 images, 9,000 training records, and RL training data from 7,713 source images. According to the release information, the best among 28 models scored 61.08 on ZoomGainBench, and ZoomGain-9B performs better on multiple fine-grained and general perception benchmarks. There are currently no further developments.
AI-generated from public reports
Developments
1 confirmed developments
- Source:English text is being prepared.
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 11
Ant Group inclusionAI: New GitHub RepositoryFirst-party
Ant Group inclusionAI Open-Sources ZoomGain: Building Supervision for Vision-Language Models via Global-Zoom Perception Gap
Ant Group inclusionAI has open-sourced ZoomGain on GitHub, using visual details that only become clear after zooming in to generate questions about the whole image, thereby building supervision signals for vision-language models. The best among 28 models scored 61.08 on ZoomGainBench, and ZoomGain-9B performs better on multiple fine-grained and general perception benchmarks. It simultaneously released an evaluation benchmark with 1,000 questions and 941 images, as well as RL training data with 9,000 training records and 7,713 source images, and 4B/9B model checkpoints.