A text-to-image model derived from NVIDIA's physical-AI foundation model family "Cosmos 3" has landed 8th among open-weight models (49th overall) on the human-vote-based Text-to-Image Arena. According to Arena's tally, the 64B-parameter Cosmos-3-Super-Text2Image was rated on par with Flux-2-Klein-9B and Qwen Image Prompt Extend, while an Agentic variant that inserts additional reasoning steps placed 11th (54th overall).
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.