Today for AI
HOT RADAR
llmHEAT 11.7°

AI CURATED HOT STORY · 10/17/2026

Elorian Launches Glimpse Model with Interactive Visual Reasoning via Scratch Space

1 reports archived1 independent sourcesupdated 10/17/2026, 00:01:37
Synthesis & Latest Updates
1 Sources Cross-Validated

Startup Elorian has released its first multimodal model, Glimpse, to address the limitations of traditional models that encode images once and reason purely in text. The model introduces a 'visual scratch space' mechanism, allowing it to dynamically crop image regions, extract video frames, or run code for measurement during reasoning, enabling iterative deep understanding of visual information. It is currently available as a research preview with API access.

LATEST/Startup Elorian has released its first multimodal model, Glimpse, to address the limitations of traditional models that encode images once and reason purely in text. The model introduces a 'visual scratch space' mechanism, allowing it to dynamically crop image regions, extract video frames, or run code for measurement during reasoning, enabling iterative deep understanding of visual information. It is currently available as a research preview with API access.

HEAT TRENDHourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. TechCrunch AIT2·68 pts
    • Glimpse uses a 'visual scratch space' to re-examine raw visual data (cropping, frame extraction, code measurement) mid-reasoning.
    • Unlike traditional multimodal models that convert images to tokens and rely on text context, Glimpse emphasizes real-time interaction between vision and logic.
    • Elorian, founded by ex-Google DeepMind researchers, raised $55M at a $300M valuation, focusing on the visual reasoning niche.