Palo Alto-based AI startup Elorian has released Glimpse, its first visual reasoning model, as a research preview with API access. Led by former Google DeepMind researcher Andrew Dai, Elorian previously raised a $55 million seed round at a $300 million valuation from investors including Striker Venture Partners, Menlo Ventures, Altimeter Capital, Nvidia, and Jeff Dean.

Unlike traditional multimodal models that convert images to text tokens once, Glimpse utilizes a ‘visual scratch space’ mechanism. This allows the system to repeatedly reference the source image, crop specific regions, pull video frames, or execute code during its internal reasoning process. Trained on a few hundred GPUs, the model is priced at $0.60 per million input tokens and $3 per million output tokens.

According to Elorian, Glimpse outperforms larger frontier models like GPT-6 Astra on scientific chart reasoning and short-form video captioning. The startup is targeting industries with extensive video assets—such as robotics, labs, and media—with the long-term goal of building models that can directly control physical systems.

Why it matters

  • Offers a cost-effective API option for enterprises needing complex visual reasoning and short-form video processing without paying top-tier frontier pricing.

  • Demonstrates that specialized, mid-sized architectures trained on small GPU clusters can outperform massive models on specific vision benchmarks.

  • Paves the way for advanced spatial reasoning tools necessary for physical robotics and automated scientific media analysis.

Source: techcrunch.com