IBM Research

📣 I’m excited to share our latest update to IBM’s family of enterprise-grade AI models: Granite 4.1. 📣 At the core of this release are the new 3B, 8B, and 30B language models. By prioritizing data quality and refinement over raw data volume, these models deliver state-of-the-art performance in tool-calling and instruction-following with predictable latency and lower token costs for enterprise users. Beyond language, the release includes: 🔎 Granite Vision 4.1: Specifically optimized for document understanding, outperforming much larger, frontier models in table and chart extraction. 🗣️ Granite Speech 4.1: High-accuracy transcription designed for noisy, real-world environments 🦾 Granite Guardian 4.1: A dedicated model for risk and policy compliance that provides calibrated uncertainty scores. 🔢 Granite Embedding R2: Scaling retrieval support to 200+ languages with a 512K context window. You can explore these models yourself on variety of platforms, including AnythingLLM, Artificial Analysis, Hugging Face, LM Studio, Ollama, OpenRouter, Replicate, Unsloth, watsonx, and Weights & Biases. IBM Research Blog URL: https://lnkd.in/dJFKCQwA

Speed wins races. Intelligence wins the future. 🏎️ IBM and Dallara are accelerating vehicle design through the development of a physics-based AI model. Early tests show how these models can evaluate high performance vehicle designs in seconds instead of hours while staying grounded in real engineering principles, and quantum exploration is already underway. Discover how this collaboration is reshaping engineering, from the racetrack to the road: https://ibm.co/6043EKSS5

Massachusetts Institute of Technology and IBM today unveiled an expanded collaboration to explore the powerful convergence of AI, algorithms, and quantum computing. The MIT-IBM Computing Research Lab will serve as a premier global hub for: 🔹 Hybrid Quantum & Classical Systems 🔹 Mathematical & Algorithmic Foundations 🔹 Foundational AI Research See how we're advancing discovery by building what's next: https://ibm.co/6042EKUPE

New week, new Circuit Breaker. 💡 Here's what stood out: an AI foundation model is helping fusion researchers push past longstanding challenges, quantum‑centric computing is coming to life at UIUC, and the latest updates from IBM Quantum this year. Catch the full breakdown 👇

AI doesn't just need new data. It needs better access to the data you already have. We’re unlocking decades of unstructured enterprise data so AI can search and understand it securely, without having to move it to the cloud. See how a 100B‑vector database fits on a single server and turns existing storage into something AI can actually use: https://ibm.co/6045EJI01

Welcome back to another week of Circuit Breaker. 👋 In this edition, we’re zooming out and drilling down: AI models learning to reason better, quantum‑centric computing taking shape with UIUC, Q4Bio finalists putting quantum to work in biology, and infrastructure updates making enterprise AI more usable at scale. Let’s break it down ↓

We’re seeing a move toward AI becoming more usable, more accountable, and more enterprise-ready. ⚡️ New ways of thinking about AI are emerging as we see how it’s being designed through the collaboration between IBM Research and ETH Zurich, how it’s being documented through the idea of an AI bill of materials, and how it’s being applied to complex, real-world data with Granite time series models. Get into the full breakdown in this week’s Circuit Breaker ↓

IBM 🤝 ETH Zürich. See what’s ahead as part of this 10-year collaboration redefining how algorithms power AI and quantum. 👇 🔹 New methods for solving problems in optimization and differential equations. 🔹 Innovations in linear algebra and Hamiltonian simulations. 🔹 New algorithmic foundations for breakthroughs across AI and quantum computing. Hear what the Director of IBM Research Zurich Alessandro Curioni has to say about the collaboration and what lies ahead: https://ibm.co/6048Eyx7u

🚀 Excited to share that Granite 4.0 3B Vision is now live! A compact vision-language model built for enterprise document understanding, with a focus on what actually matters in production: extracting structured data from complex visual documents — tables, charts, and key-value pairs. Despite its size, it delivers excelent performance across extraction tasks. 💡 What we’re particularly excited about: * Modular by design - Delivered as a LoRA adapter on top of Granite 4.0 Micro, enabling a single deployment to handle both multimodal and text-only workloads seamlessly. * Deep visual reasoning - A DeepStack-style architecture that injects both high-level semantics and fine-grained spatial details — critical when layout matters as much as content. * ChartNet dataset - A million-scale dataset for chart understanding, enabling the model to move beyond description to true structured reasoning. 🧩 Where it fits Designed to plug directly into real pipelines — either as a standalone extraction model or as part of a full document understanding stack with #Docling . Think: invoices, reports, forms, scientific documents — turning messy visuals into structured, usable data. Huge credit to the team at #IBMResearch (#IBMResearchIsrael and our collaborators at MIT-IBM). Rogerio Feris, Artem Spector, Avihu Dekel, Ben Wiesel, Dhiraj Joshi, Hang H., Isaac Sanchez, Jovana Kondić, Moshe Kimhi, Nimrod Shabtay, Pengyuan Li, Roei Herzig, Shafiq Abedin, Shaked Perek, Sivan Harary, Madison Lee, Abraham Daniels, Kate Soule, Udi Barzelay, Peter W. J. Staar, Aya Soffer, Shila Ofek-Koifman, Tal Drory, Luis Lastras, David Cox 👉 Available now on Hugging Face (Apache 2.0). Would love to see what you build with it. https://lnkd.in/dyswimUe #AI #MultimodalAI #VisionLanguageModels #DocumentAI #IBM

This week felt like a turning point 🚀 Quantum simulations are aligning with real lab data, AI is capturing patient notes even in noisy mid‑flight conditions, momentum is building around standardizing how we scale LLMs, and a clear takeaway from All Things AI is emerging: smaller models are having a moment and orchestration is quickly becoming the name of the game. Catch up on everything in this week’s Circuit Breaker ↓