Elorian Launches Glimpse Visual Reasoning Model
Palo Alto startup Elorian has launched its first AI model, Glimpse, a visual reasoning system that outperforms larger models on specific tasks.
Elorian, a Palo Alto AI startup, announced its first model, Glimpse, and secured $55 million in seed funding at a $300 million valuation. The company, founded six months ago by ex-DeepMind researcher Andrew Dai, came out of stealth with backing from Striker Venture Partners, Menlo Ventures, Altimeter Capital, Nvidia, and Google chief scientist Jeff Dean.
Elorian released Glimpse on Friday as a research preview with API access and a waitlist-only chat interface. The model is designed to reason with images by maintaining access to the source visual data, allowing it to revisit images and video during reasoning, crop details, or even execute code to examine what it sees. This dynamic interaction challenges the industry's standard static image-to-text processing.
Glimpse model details and pricing
Glimpse is available as a research preview with the following pricing structure:
| Metric | Cost |
|---|---|
| Input tokens | $0.60 per million |
| Output tokens | $3 per million |
Technical capabilities and performance
Glimpse outperforms OpenAI's GPT-6 Astra on benchmarks for scientific chart reasoning and short-form video captioning. This performance is achieved despite the model being substantially smaller than frontier systems. Elorian built Glimpse with a small team and a few hundred GPUs.
Founder Andrew Dai highlighted the core challenge the model addresses. "Humans can look at Ikea diagrams, with no text, and assemble Ikea furniture, right? Why can’t AI models?" he said. The model's ability to dynamically interact with visual data, such as cropping or selecting frames mid-reasoning, is central to its approach.
Target customers and strategic vision
The initial target customers include robotics firms, scientific labs, and media and creative companies that possess large video datasets. Elorian aims to build what it calls 'the world's best visual reasoning model' by 2027, leading on every visual reasoning test. The long-term vision is to enable direct robot control through advanced scene understanding, targeting what the company terms visual AGI.
This launch arrives as competitors like OpenAI have also added capabilities for models to crop and zoom during reasoning. Elorian's goal is to lead the field in visual reasoning by the end of the decade.
