Elorian
Elorian is an AI research and product lab building models that natively understand and reason through visual information. Its stated goal is to give AI grounded visual thinking about spatial relationships, physical constraints, design intent and abstractions so it can solve complex problems in the physical world.
At a Glance
- Robotics companies
- Scientific laboratories and research organizations
- Media and video companies
- Creative companies
- +3 more
AI Tools by Elorian
(1)Elorian
Visual Reasoning Foundation Models
Discussions
No discussions yet
Be the first to start a discussion about Elorian
Latest News
Elorian introduced ScienceVision, a 250-question scientific 3D-artifact reasoning benchmark built with UniPat AI.
Elorian released Glimpse, its first visual-thinking model, with API access and a chat interface.
Elorian published its Glimpse technical/product announcement, describing visual scratch space, iterative inspection and benchmark results.
Elorian and Scale AI launched Humanity's Sixth Sense (HSS), a benchmark for implicit visual reasoning.
Products & Services
Elorian's first visual-thinking foundation model, available through an API and chat UI. It reasons about the physical world, revisits visual evidence during inference, follows objects through motion and changing viewpoints, understands spatial relationships, uses multimodal tools, and can coordinate subagents.
Hosted developer and chat platform at platform.elorian.ai, exposing the Glimpse model through an OpenAI-compatible API at api.elorian.ai/v1, with image and video inputs, streaming, file uploads, background runs and generated-file downloads.
A benchmark of 522 image and short-video tasks measuring implicit visual reasoning across temporal/causal dynamics, physical/spatial logic, social understanding and abstract/contextual inference; built with Scale AI.
A benchmark of 250 expert-level questions across biology, chemistry, medical imaging, mechanical CAD and architecture, testing reasoning over scientific 3D artifacts and fixed views; built with UniPat AI.
Market Position
Elorian positions Glimpse against general frontier multimodal models from OpenAI, Anthropic and Google by making visual representations part of the reasoning loop rather than primarily converting an image to text and reasoning over that description. It emphasizes grounded, iterative visual inspection, while claiming frontier-level benchmark performance at substantially lower cost and with a smaller model/team. The practical competitive set includes GPT multimodal models, Claude multimodal models, Gemini and other visual-reasoning/agent platforms.
Leadership
Founders
Andrew Dai
Co-founder and CEO. Former Google Brain and Google DeepMind researcher; developed foundational techniques for language-model pretraining and supervised fine-tuning, co-led early mixture-of-experts work including GLaM, and led Gemini data and co-led PaLM 2 pretraining.
Yinfei Yang
Co-founder and Chief Multimodal Architect. Formerly at Apple and Google Research; led Apple Foundation Model multimodal work and the MM1/MM1.5 model family, and co-created Google's ALIGN contrastive language-image pretraining model.
Seth Neel
Co-founder and Member of Technical Staff. Former Harvard professor researching data-centric generative AI, formerly with Google Research/Gemini Data, and co-founder/CTO of a machine-learning energy startup.
Executive Team
Andrew Dai
Co-Founder & Chief Executive Officer
Former Google Brain and DeepMind researcher; worked on language-model pretraining, GLaM, Gemini data and PaLM 2 pretraining.
Yinfei Yang
Co-Founder & Chief Multimodal Architect
Former Apple and Google Research researcher; led Apple's multimodal foundation-model work and co-created ALIGN.
Founding Story
Elorian was started to address what its founders saw as a fundamental weakness in contemporary AI: vision-language systems commonly translate visual input into text and then reason over that description, which is fragile for spatial, physical and implicit visual tasks. The founders' initial vision was a foundation for visual thinking that reasons in visual space and ultimately connects perception to action in the physical world.
Business Model
Revenue Model
Usage-based API credits for developers, with enterprise volume access, custom evaluation, deployment support and procurement assistance. The hosted chat platform also uses account plans/credits; enterprise pricing is negotiated by contract.
Pricing Tiers
API requests consume credits from the first token.
Published API output-token rate.
Enterprise limits and prices are agreed by contract; Elorian also offers volume access, custom evaluation and deployment support.
Target Markets
- Robotics companies
- Scientific laboratories and research organizations
- Media and video companies
- Creative companies
- Architecture and design organizations
- Automotive companies
- Robotics and embodied systems
- Scientific discovery and laboratory-data analysis
- Architecture and design
- Automotive and physical-world applications
- Creative industries and media/video analysis
- Social-media video tagging and captioning