Elorian, the Palo Alto startup that raised a $55 million seed round at a $300 million valuation before it had a product, has released its first model. It’s called Glimpse, and it went live Friday as a research preview with API access and a chat interface, TechCrunch AI reports. Elorian’s argument is that today’s AI models are very good at language and code but still can’t really see.
The company’s fix is to let the model look at an image more than once.
Why AI still struggles with images
Most multimodal models look at an image once, turn it into tokens, and then reason about it in text. It’s a bit like describing a photo from memory. Glimpse can keep going back to the image while it reasons.
Elorian CEO Andrew Dai, a former Google DeepMind researcher, saw the problem firsthand on a game night. He photographed a board game he’d just finished with friends and asked a frontier model who was winning. It couldn’t tell him.
Board games “each have their own visual language, each have their own iconography, symbols,” Dai told TechCrunch. “To an adult, it’s quite easy to understand what’s going on with minimal text, but these models aren’t able to intuit what the symbols mean.” He also put it another way: “Humans can look at Ikea diagrams, with no text, and assemble Ikea furniture, right? Why can’t AI models?”
5 things to know about Glimpse
- It has a “visual scratch space.” While it’s reasoning, Glimpse can crop into part of an image, pull specific frames from a video, or run code to measure and compare what it sees. Think of pinching to zoom on a phone photo to check what you thought you saw.
- It’s small and was cheap to train. Elorian says a small team trained Glimpse on a few hundred GPUs. The company also says the model is much smaller than the frontier systems it’s benchmarked against.
- The pricing lands in the middle. Glimpse costs $0.60 per million input tokens and $3 per million output tokens. That’s cheaper than mid-level models like Sonnet and Sol, but more than lightweight models like Haiku and Luna.
- It posts some strong benchmark results. In Elorian’s own evaluations, Glimpse beats GPT-6 Astra on scientific chart reasoning and short-form video captioning. On other visual reasoning tasks, its results are competitive rather than leading.
- It’s built for companies with lots of video. Elorian’s first targets are robotics firms, scientific labs, media companies, and creative businesses. These teams have huge visual archives, and a model that can actually inspect frames matters to them.
How it compares to what’s already out there
The core idea isn’t new, and Elorian doesn’t say it is. OpenAI added crop-and-zoom during reasoning to its o-series models in 2025. Elorian’s claim is that it built the whole model around that ability, and did it cheaply.
That’s the interesting part. Frontier labs added visual inspection to models designed mainly for text. Elorian started from the other end. If a smaller, purpose-built model can beat much larger ones on chart reading and video captioning, that’s a sign that architecture choices still matter more than raw scale for specialized tasks.
Caveats worth noting
- It’s a research preview. Expect rough edges and changes before any general release.
- The benchmarks are self-reported. The comparison with GPT-6 Astra comes from Elorian’s own evaluations. Independent testing will show whether the gains hold up.
- Its lead is narrow. Glimpse leads on specific tasks, not across the board.
The backers and the bigger plan
Elorian came out of stealth six months ago. Its round was co-led by Striker Venture Partners, Menlo Ventures, and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean. For a company this young, that’s a strong group of backers.
Elorian’s plans go beyond image analysis. Over the long term it wants to go from seeing to doing: building a model of a scene, using it to choose an action, and eventually controlling robots directly. That ties Glimpse to the wider push toward world models and embodied AI, where seeing accurately comes before acting.
The near-term goal is ambitious too. Next year, Dai says, Elorian aims to build “the world’s best visual reasoning model,” one that leads on every visual reasoning test. Whether a small team on a few hundred GPUs can pull that off against the big labs is the question to watch. You’ll find more details in the original TechCrunch AI report.