An open-source 8B model getting ~64% of Nano-Banana-Pro on infographic benchmarks is not nothing
Most T2I models can make a nice-looking image.
Far fewer can make a readable infographic.
SenseNova just released SenseNova-U1-8B-MoT-Infographic, an open 8B model tuned for dense visual documents: labels, layouts, charts, posters, explainer pages, small text blocks.
The numbers are weird enough to be worth testing.
Using a rough composite of BizGenEval + IGenBench, it gets to about 64% of Nano-Banana-Pro’s level. More interestingly, it comes out slightly above GPT-Image-1.5 on that same rough average.
On BizGenEval hard split:
- SenseNova-U1-8B-Infographic: 46.6
- GPT-Image-1.5: 35.9
It is obviously not a solved problem. Infographics are brutal. But this is the first open 8B checkpoint I’ve seen that looks specifically aimed at the boring stuff people actually need: readable diagrams and visual explanations.
Showcases:
https://github.com/OpenSenseNova/SenseNova-U1/blob/main/docs/u1_infographic_showcases.md
Github Repo:
https://github.com/OpenSenseNova/SenseNova-U1
Discord: