Dispatch
Luma Uni-1: Autoregressive Reasoning Image Model Beats Google Nano Banana 2 and GPT Image 1.5 at 30% Lower Cost
Luma Labs released Uni-1, a decoder-only autoregressive transformer that reasons through spatial relationships and logical constraints before rendering images. Unlike diffusion models, Uni-1 treats text and images as interleaved token sequences. It tops Google's Nano Banana 2 and OpenAI's GPT Image 1.5 on reasoning-based benchmarks, nearly matches Gemini 3 Pro on object detection, and costs ~$0.09 per 2048px image — 10-30% cheaper than competitors.
Source
↳ Follow the thread