Phi-4-reasoning-vision-15B Technical Report

Researchers released Phi-4-reasoning-vision-15B, a compact open-weight multimodal AI model that combines vision and language capabilities with strong performance in scientific and mathematical reasoning. The model demonstrates that careful architecture design and high-quality data curation can enable smaller models to achieve competitive performance with less computational resources.

• →Phi-4-reasoning-vision-15B is a 15 billion parameter open-weight multimodal model optimized for vision, language, and reasoning tasks.
• →The model excels at scientific and mathematical reasoning while maintaining efficiency in training and inference compute requirements.
• →Data quality through systematic filtering, error correction, and synthetic augmentation proved to be the primary driver of model performance.
• →High-resolution dynamic-resolution encoders provide consistent improvements for accurate perception and reasoning capabilities.
• →The hybrid approach uses mode tokens to switch between fast direct answers for simple tasks and chain-of-thought reasoning for complex problems.