Ideogram has open-weighted its 4.0 image model: a 9.3-billion-parameter single-stream diffusion transformer, its first downloadable foundation model, with nf4 and fp8 weights on Hugging Face.

The interesting part is the size mismatch. Ideogram reports that the sub-10B model beats FLUX.2 [dev] at 32B and HunyuanImage 3.0 at 80B on text rendering, and it ranks first among open-weight models on DesignArena. Text and image tokens share the same projections across all 34 layers, and prompts can be written as structured JSON that pins layout, colour and text placement instead of a paragraph of prose, aimed squarely at design work where legible, correctly placed text usually breaks.

The catch is the licence. The inference code is Apache 2.0, but the weights ship under a non-commercial agreement: free to download, fine-tune and run for research, with production use requiring a separate paid deal.