With open-source availability and integrations across major platforms, these AI solutions offer a powerful toolkit for developers, enterprises, and consultants, paving the way for future AI agents and innovations in business operations.
IBM has announced the launch of Granite 3.0, its latest suite of AI models. This release introduces powerful language models and specialized tools, all available under the open Apache 2.0 license, setting a new benchmark for enterprise AI performance, safety, and flexibility.
Granite 3.0 8B and 2B models excel in core AI tasks like Retrieval Augmented Generation (RAG), classification, and summarization. Mixture-of-Experts (MoE) models offer efficient inference with low latency, optimized for CPU-based deployments and edge computing. The Granite Guardian 3.0 models deliver the company’s most advanced safety features to detect risks such as bias, toxicity, and hallucinations. These models outperform Meta’s Llama Guard models across several safety benchmarks.
New tools for developers through the watsonx platform:
- Next-Gen Coding Assistant: Powered by Granite models, the updated watsonx Code Assistant now supports multiple programming languages, including C++, Java, and Python.
- AI Agent Capabilities: The Granite 3.0 models introduce new agentic features, enabling advanced reasoning and multi-step problem-solving. Future releases will enhance AI agents for domains like cloud transformation and business operations.
The updated time-series models achieve state-of-the-art results in zero/few-shot forecasting, surpassing models 10 times larger from Google and Alibaba. The models will power the company’s Consulting Advantage, boosting productivity for it’s 160,000 consultants. The models are integrated into platforms like Hugging Face, NVIDIA NIM microservices, and Google Vertex AI, supporting edge deployments through tools like Ollama and Replicate. The models cements the company’s commitment to advancing trustworthy AI, providing scalable, open-source solutions that meet enterprise needs efficiently and responsibly.
For more information, click here.
