IBM’s newest AI model, Granite Vision 3.3 2B, earned second place on the OCRBench leaderboard, which made it the best-performing multimodal model with fewer than 7 billion parameters. This is part of IBM’s bigger plan to develop AI models that understand the world through multiple senses. Indeed, Granite Vision is designed to understand visual data such as charts, tables, handwritten notes, and images in documents, which are the types of things many people interact with at work every day. This comes after its new speech recognition model also showed impressive results.
Don’t Miss TipRanks’ Half-Year Sale
To test how well these models understand visual content, the AI industry uses a benchmark called OCRBench. It measures how accurately a model can read text, recognize key information, understand math written by hand, and answer detailed questions about what it sees. Each model is tested with 1,000 questions, where every answer must contain at least four symbols to make sure the results are meaningful. Granite Vision 3.3 2B performed especially well in reading handwritten math and answering scene-related questions. It even outperformed larger models from companies like Google (GOOGL), OpenAI, and Meta (META), which demonstrates just how effective IBM’s training approach was.
Interestingly, IBM made key updates to this version of Granite Vision. In fact, the team added a new encoder and trained the model using more document-focused examples that were specifically chosen to match IBM’s business needs. Researcher Rogerio Feris explained that the model is not only powerful but also small and efficient, therefore making it easier and cheaper to use. Looking ahead, the IBM team believes that models like Granite Vision could begin to reason and act on visual information without detailed instructions.
What Is the Target Price for IBM?
Turning to Wall Street, analysts have a Moderate Buy consensus rating on IBM stock based on seven Buys, five Holds, and two Sells assigned in the past three months, as indicated by the graphic below. Furthermore, the average IBM price target of $269.46 per share implies 8.6% downside risk.
See more IBM analyst ratings
Disclaimer & DisclosureReport an Issue