OmniReasoningBench introduces 1,150 questions requiring both audio and visual evidence for correct answers. It is supported by the OmniQA data engine, which automatically constructs evidence-grounded QA pairs to elicit joint modality reasoning.
HOW THIS AFFECTS YOU
●
builderThis allows for more rigorous testing of omni-modal features in your products.
●
researcherYou can now benchmark whether your multimodal models actually reason across modalities rather than just one.