IQ1-V

A 7B-parameter vision-language model engineered for precision document parsing and complex spatial reasoning.

IQ1-V pairs a high-resolution visual encoder with a 7B-parameter backbone to solve complex multimodal reasoning. It dominates document-heavy benchmarks (like DocVQA) and maintains a high score on the MMMU leaderboard. The system processes inputs at 1024x1024 resolution: this allows for granular extraction from blueprints, maps, and financial tables. It is built for speed and accuracy (delivering enterprise-grade performance in a compact footprint).

https://huggingface.co/IQ-AI/IQ1-V