Elorian Releases Its First AI Model, Glimpse
Palo Alto artificial intelligence startup Elorian has officially released its first model, Glimpse, as a research preview featuring API access and a chat interface.
Radwan AbdulhadiEditor and technology writer
Introduction to Glimpse
Palo Alto AI startup Elorian, which previously raised a $55 million seed round at a $300 million valuation prior to launching any product, has released its first model, according to a report from TechCrunch. The model, named Glimpse, is launching as a research preview that includes both API access and a chat interface. According to the company, today's artificial intelligence systems still struggle to truly see, despite possessing advanced skills in language and code.
Most conventional multimodal models analyze an image once, translate it into tokens, and subsequently reason about it purely in text, a process similar to describing a photo entirely from memory. In contrast, Glimpse continually references the source material. During the reasoning process, the model can crop into a specific region, extract particular frames from a video, or execute code to measure and compare visual details. Elorian refers to this unique mechanism as a 'visual scratch space,' likening it to a human pinching and zooming on a smartphone photograph to verify details.
The Inspiration Behind Visual Reasoning
Andrew Dai, CEO of Elorian and a former researcher at Google DeepMind, noted that one of the primary indicators of persistent limitations in computer vision occurred during a casual game night. Last year, Dai photographed a board game he had just completed alongside friends and asked a frontier model which player was winning. The model was unable to provide an answer.
Dai explained that board games possess unique visual languages, iconography, and symbols. While an adult can easily interpret the state of a game with minimal textual information, current artificial intelligence models struggle to intuit the meanings behind those visual symbols. He further noted that humans can easily look at diagrams without any text and assemble furniture, raising the question of why models should not be capable of doing the same.
Training Approach and Cost
While the core concept is not entirely unprecedented—OpenAI integrated crop-and-zoom capabilities during reasoning into its o-series models in 2025—Elorian emphasizes that it built its entire architecture around this capability in a cost-effective manner. The startup stated that a small team successfully trained Glimpse utilizing only a few hundred GPUs, resulting in a model that is substantially smaller than the larger frontier systems against which it has been benchmarked.
Regarding pricing, Glimpse costs $0.60 per million input tokens and $3 per million output tokens. This places the model at a price point that is cheaper than mid-level alternatives like Sonnet and Sol, though it remains more expensive than lightweight offerings such as Haiku and Luna.
Performance and Future Goals
Based on internal evaluations conducted by Elorian, Glimpse outperforms even GPT-6 Astra on scientific chart reasoning tasks and short-form video captioning. Furthermore, the model demonstrates competitive results across various other visual reasoning benchmarks.
Elorian initially emerged from stealth six months ago supported by a funding round co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with additional participation from Nvidia and Google chief scientist Jeff Dean. The startup's initial market targets include organizations managing large volumes of video data, such as robotics firms, scientific laboratories, media organizations, and creative companies.
In the long term, Elorian aims to transition its technology from observation to action. This roadmap involves constructing a model of a physical scene and utilizing that understanding to select and execute actions, including the direct control of robots. Looking ahead to the following year, CEO Andrew Dai stated that the company intends to build the premier visual reasoning model globally, targeting top performance across every visual reasoning test.
Sources
- TechCrunchElorian’s first AI model wants to reason with images, not just about them
Continue chronologically
Related entity coverage
Related stories
AI Industry Divided Over Future of Regulatory Frameworks
Tech leaders clash over AI safety proposals as the Trump administration labels industry-led regulation efforts a hoax amidst internal industry conflict.
UK Launches $134M AI Biosecurity Lab for Safety
The UK government has established a new $134 million AI Biosecurity Lab backed by major tech firms to detect biological threats and accelerate vaccines.
OpenAI, Anthropic, and Nvidia Hit Record Lobbying Quarters
Leading artificial intelligence companies ramped up their federal lobbying spending ahead of the midterms, according to new federal disclosures.
OpenAI and Anthropic Urge UK Lawmakers to Regulate AI
Representatives from OpenAI and Anthropic told a British parliamentary hearing that voluntary industry pledges are no longer enough for safety.
