Glimpse

VISUAL THINKING

Elorian's first visual-thinking foundation model: reasons from pixels via a visual scratch space (crops into regions, rewatches intervals at denser fps, runs measuring code) instead of describing images then reasoning in text; beats or matches GPT-6 Astra on scientific chart reasoning (CharXiv-R 92.9) and short-form video captioning (UGC-VideoCap 82.66) at $0.60/$3.00 per 1M tokens - 94% below Astra; launched Oct 16, 2026 as a research preview by the ex-DeepMind Palo Alto startup.

Context

262,144 tokens

Architecture

Undisclosed (substantially smaller than frontier models; post-trained on an open-weights base)

Official API (1M)

$0.6 in / $3 out

License

Elorian API (research preview)

Quickstart

curl -s https://modelregistry.tirup.in/api/cli?model=glimpse

OFFICIAL ANNOUNCEMENT

BACK TO LEDGER

Sources

Introducing Glimpse: visual thinking that stays grounded · Elorian
Elorian's first AI model wants to reason with images, not just about them · TechCrunch
Glimpse listing and pricing · OpenRouter

Report an error in this record