SperidLabs' Iris-Z: One Unified Generative Model Beats Specialist Depth Models at Scale

speridlabs · hf · 2026-11-17

SperidLabs released the technical report for Iris-Z, arguing that the bitter lesson applies to computer vision and 3D/spatial understanding just as it did to language: effort should go into unified architectures that scale with data, parameters, and compute rather than task-specific designs.

Key points:

Authored by Alejandro Cañada and Chema Garabito, positioned as an empirical test of the generalist-model path toward visual intelligence.

Original post →

More from Multimodal

Multimodal channel →