Breakthrough AI Mimics Physics, Costs 90% Less Than Rivals

Perceptron Inc. has introduced Mk1, a new AI model designed to understand the complexities of the physical world through video analysis. This model goes beyond traditional image recognition by interpreting cause-and-effect, object dynamics, and the laws of physics. It achieves impressive scores on spatial and video benchmarks, surpassing rivals like Google's Gemini and OpenAI's GPT-5. For instance, Mk1 achieved 85.1 on EmbSpatialBench and 88.5 on VSI-Bench.

The flagship Mk1 model is priced at $0.15 per million input tokens and $1.50 per million output tokens. This pricing is substantially lower, up to 90% less than competitors, positioning Mk1 on the 'Efficiency Frontier' for accessible, high-performance AI. The technology can process video natively up to 2 frames per second with a 32K token context window, maintaining temporal continuity and object identity.

Mk1's unique 'Physical Reasoning' capability allows it to understand object dynamics and physical interactions, distinguishing it from other models. Perceptron also offers an expanded developer platform with specialized functions like 'Focus' and 'Counting.' In parallel, the company maintains its open-weights 'Isaac' series for edge deployments, providing flexible licensing strategies for diverse use cases.

Founded by former Meta AI researchers, Perceptron aims to make 'physical AI' as ubiquitous as digital AI. Early adopters are already leveraging Mk1 for applications such as automatic sports highlight clipping, robotics data curation, and manufacturing quality control. The company envisions a future where AI seamlessly integrates with the physical world.