Inside Iris-1: our CEO and engineers on our first world model

Marc Gyöngyösi, Founder and CEO 0:02
Look around you. Almost everything you have in your house has at one point or another been in a warehouse, a factory, on a truck on the road, maybe even in a field on a farm. While agentic AI is changing everything about the digital world, businesses operating in the physical world are still left behind because AI agents fundamentally don't know what happens on the floor. Today, we are changing that. We're excited to introduce Iris-1.

Rohan Patel, AI Software Engineer 0:37
Iris-1 is a world model. It has watched more warehouse safety events than a human could possibly watch, and it natively understands concepts like people, pallets, forklifts, and racks.

Marc Gyöngyösi 0:49
In a world where every second matters and lives are at stake, you cannot trade accuracy for cost.

William Wang, VP of Software Engineering 0:56
After nearly a decade of deploying vision AI in logistics and manufacturing, we've learned what clients really need: solutions that are simple, sharp, and pragmatic.

Rohan Patel 1:06
Comparing Iris-1 head-to-head with other models, you'll notice it provides a more clear, succinct description than the other models. It describes occupancy, motion, machine type, common events, and PPE, just the way that a warehouse expert would describe it. The performance gap over the VLMs is particularly pronounced when it comes to motion and dynamics, such as the lift-on-lift collision it identifies correctly here, which the frontier models did not.

William Wang 1:34
We are already teaching it to see and understand more. And for our clients, that means full visibility into their operation and a fraction of the cost.

Rohan Patel 1:43
Iris-1 can run on consumer-grade GPUs rather than in a data center. After watching the video once, asking Iris follow-up questions about the event costs almost nothing. On a held-out test set, Iris-1 outperforms three frontier models on describing and detecting collisions, while also being cheaper and faster.

Marc Gyöngyösi 2:05
Consider, for example, the safety leader of a national 3PL. In the past, the safety leader would have to watch hundreds of video clips to understand what happens across their 40-plus sites. Today with Iris, Iris does all the watching for them.

William Wang 2:21
That's the philosophy behind everything we built, and why Iris-1 is effective and built to scale. It shows operators where to focus, so they can stop reacting and start improving, across their sites and the whole network.

Marc Gyöngyösi 2:38
While Iris-1 is a big step, it's also a small first step, and we're excited about what future iterations of Iris will be able to do for our clients well beyond the four walls of warehouses, all the way into factories, yards, and farms.