Today for AI
HOT RADAR
llmHEAT 6.0°

AI CURATED HOT STORY · 10/14/2026

Apple and Duke Introduce LensVLM: Long-Context Image Compression with On-Demand Zooming for NeurIPS 2026

1 reports archived1 independent sourcesupdated 10/14/2026, 16:01:00
Synthesis & Latest Updates
1 Sources Cross-Validated

Apple's Foundation Model team and Duke University introduced LensVLM, which compresses long contexts into image pages and enables on-demand zooming to retrieve details. The method achieved 68.9% average accuracy across seven text QA benchmarks with 4.3x effective compression, approaching the 72.4% baseline of full-text input, and was accepted at NeurIPS 2026.

LATEST/Apple's Foundation Model team and Duke University introduced LensVLM, which compresses long contexts into image pages and enables on-demand zooming to retrieve details. The method achieved 68.9% average accuracy across seven text QA benchmarks with 4.3x effective compression, approaching the 72.4% baseline of full-text input, and was accepted at NeurIPS 2026.

HEAT TRENDHourly heat curve

3 fully observed hours
Now
6.0
Peak
6.320:00
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. 机器之心 (微信公众号)T2·78 pts
    • LensVLM converts long text into compressed images, allowing models to skim first and then zoom in for details like reading a document.
    • Achieved 68.9% average accuracy with 4.3x compression, only 3.5 percentage points below the uncompressed baseline.
    • Solves the irreversible loss of key details in traditional 'one-way' compression schemes by enabling on-demand retrieval of original content.
Apple and Duke Introduce LensVLM: Long-Context Image Compression with On-Demand Zooming for NeurIPS 2026 | Today for AI