MMaDA

[junior plume]

Multimodal reasoning diffusion LLM, based on LLaDA
paper | github | hf | demo

It can read and generate images and text

Also, all of MMaDA's training data is open (but it initializes from LLaDA which has closed training data)

afaik this is the first open reasoning model that can do image generation

[sharp veldt]

Bytedance actually released one yesterday, BAGEL: https://demo.bagel-ai.org/

[junior plume]

oh, I saw BAGEL but didn't see that it had reasoning. Nice!