Open, performant and trusted AI models built for business.
IBM Granite 4.0: the next generation of Granite models
Lower costs and speed up workloads with efficient models designed for scalable, enterprise-ready AI adoption.
Meet Granite
Build and scale AI faster with customizable, open-source models optimized for enterprise workloads, cost efficiency, and flexible deployments.
Open
Open source under Apache 2.0, Granite ensures transparency, while enabling full customizability and deployment flexibility across any infrastructure.
Performant
The small, high-performing models are designed to maximize efficiency and scalability for essential enterprise tasks
Trusted
Eliminate the risk of “black box” AI with transparency into training data and processes, harm detection capabilities and built-in guardrails.
Introducing Granite 4.0
Meet the models
Granite 4.0 Nano
Lightweight, local, and edge AI tasks where compute and connectivity are limited
Granite 4.0 Micro & Tiny
High-volume, low-complexity tasks where speed, cost, and efficiency are the top priority
Granite 4.0 Small
Enterprise workflows that require stronger performance without the cost of frontier models
By the numbers
70%+
reduction in memory requirements
2x
faster inferencing speeds
Granite Family of Small Models
Core language models with reasoning, optimized for RAG and agentic workflows.
Models generating high-quality text embeddings for semantic search, RAG, and contextual multi-turn information retrieval.
Ultra-compact vision-language model converting documents into structured, machine-readable formats while preserving layout, tables, and equations.
Efficient vision-language models for document and image understanding, enabling OCR, chart analysis, and enterprise content extraction.
Lightweight speech-language models for transcription and translation across 7 languages, delivering strong accuracy and efficiency.
Lightweight pre-trained models for fast, accurate time-series forecasting, optimized for efficient deployment across hardware environments.
Guardrail models detecting hallucinations, bias, harmful content, and jailbreaks, ensuring safe enterprise AI deployment across workflows.
NASA-IBM models for Earth observation, predicting biomass, climate, land temperature, and floods from large-scale satellite data.
Performance and efficiency
Granite 4.0 is engineered for efficiency, using less memory while delivering faster speeds and high performance. This balance allows enterprises to reduce costs and scale solutions faster across critical workloads.
Granite 4.0 models are designed to do more with less. They use dramatically less memory - over 70% less than similar models - so organizations can run powerful AI on more affordable hardware. That means lower infrastructure costs, faster performance, and the ability to scale AI more easily across the business.
Granite 4.0 delivers consistently high throughput as workloads scale, handling larger batch sizes with ease while other models slow down. This ensures enterprises can maintain reliable performance for applications that need to serve many users or complex tasks at once.
Granite 4.0 delivers stronger accuracy with far lower memory requirements than competing models, even at smaller sizes. That efficiency translates into cost savings, greater accessibility, and the ability to deploy enterprise AI more widely and flexibly.
Granite 4.0 outperforms both similarly sized and larger open models on RAG tasks. By delivering higher accuracy without demanding extra infrastructure, Granite helps enterprises build more reliable, knowledge-grounded applications while keeping deployments efficient and cost-effective.
Granite 4.0 demonstrates industry-leading instruction-following performance among open models, an essential capability for agentic workflows. By balancing strong accuracy with smaller size, Granite provides enterprises with high-quality outputs for complex tasks at lower infrastructure costs than larger open models.
Granite 4.0 models are designed to do more with less. They use dramatically less memory - over 70% less than similar models - so organizations can run powerful AI on more affordable hardware. That means lower infrastructure costs, faster performance, and the ability to scale AI more easily across the business.
Granite 4.0 delivers consistently high throughput as workloads scale, handling larger batch sizes with ease while other models slow down. This ensures enterprises can maintain reliable performance for applications that need to serve many users or complex tasks at once.
Granite 4.0 delivers stronger accuracy with far lower memory requirements than competing models, even at smaller sizes. That efficiency translates into cost savings, greater accessibility, and the ability to deploy enterprise AI more widely and flexibly.
Granite 4.0 outperforms both similarly sized and larger open models on RAG tasks. By delivering higher accuracy without demanding extra infrastructure, Granite helps enterprises build more reliable, knowledge-grounded applications while keeping deployments efficient and cost-effective.
Granite 4.0 demonstrates industry-leading instruction-following performance among open models, an essential capability for agentic workflows. By balancing strong accuracy with smaller size, Granite provides enterprises with high-quality outputs for complex tasks at lower infrastructure costs than larger open models.
Trusted by companies across all industries
Crushbank
"At CrushBank, we've seen first-hand how IBM's open, efficient AI models deliver real value for enterprise AI – offering the right balance of performance, cost-effectiveness, and scalability. Granite 3.2 takes it further with new reasoning capabilities, and we're excited to explore them in building new agentic solutions."
David Tan
CTO
CrushBank
Granite for developers
Build a document summarizer with IBM Granite to process documents beyond context window limits.
Build a RAG pipeline with Granite to answer queries using an external knowledge base.
Build a multimodal RAG pipeline with Granite and Docling to query text, tables, and images.
Use Granite to perform entity extraction and output clean JSON with prompt- or class-based definitions.
Use Granite time series models to perform zero-shot and fine-tuned time series forecasting.
Use automatic speech recognition (ASR) to generate a podcast transcript using Granite and watsonx.ai.
Build a local AI co-pilot using IBM Granite Code, Ollama, and Continue.
View the full granite cookbook
Build with Granite
Granite models drive the AI behind many IBM products and services. Discover ready-to-use solutions for code generation, application development, and model testing. All powered by IBM Granite.
Analysts and leaderboards
Read the report to see how IBM empowers data scientists and machine learning engineers to build, deploy, and govern impactful AI applications across their enterprises.
Discover why Granite Guardium models rank in six of the top 10 spots on the GuardBench Leaderboard, excelling at identifying harmful or malicious prompts and LLM-generated responses.
Explore the leaderboard in which IBM is ranked as the top speech models based on low word error rate.
Explore why Granite earned a top rating in the Stanford Foundation Model Transparency Index, which assesses openness around data sources, dataset size, harmful content filtering, and other key transparency factors.
Stay on top of AI news
OpenAI's gpt-oss-120B is now available in IBM watsonx.ai’s developer studio—giving you secure, flexible access to one of the most advanced open models
In episode 67 of Mixture of Experts, the panel debriefs OpenAI’s release of gpt-oss, their new open-source models.
Get curated selection of AI topics, trends and research sent directly to your inbox.
Learn how AI is poised to revolutionize the creation of beauty products, to make them even more sustainable and innovative.
IBM believes in the creation, deployment and utilization of AI models that advance innovation across the enterprise responsibly. IBM watsonx AI and data platform have an end-to-end process for building and testing foundation models and generative AI. For IBM-developed models, we search for and remove duplication, and we employ URL blocklists, filters for objectionable content and document quality, sentence splitting and tokenization techniques, all before model training.
During the data training process, we work to prevent misalignments in the model outputs and use supervised fine-tuning to enable better instruction following so that the model can be used to complete enterprise tasks via prompt engineering. We are continuing to develop the Granite models in several directions, including other modalities, industry-specific content and more data annotations for training, while also deploying regular, ongoing data protection safeguards for IBM developed models.
Given the rapidly changing generative AI technology landscape, our end-to-end processes are expected to continuously evolve and improve. As a testament to the rigor IBM puts into the development and testing of its foundation models, the company provides its standard contractual intellectual property indemnification for IBM-developed models, similar to those it provides for IBM hardware and software products.
Moreover, contrary to some other providers of large language models and consistent with the IBM standard approach on indemnification, IBM does not require its customers to indemnify IBM for a customer's use of IBM-developed models. Also, consistent with the IBM approach to its indemnification obligation, IBM does not cap its indemnification liability for the IBM-developed models.
The current watsonx models now under these protections include:
(1) Slate family of encoder-only models.
(2) Granite family of a decoder-only model.
* How smaller, industry-tailored AI models can offer greater benefits
https://www.ft.com/partnercontent/ibm/how-smaller-industry-tailored-ai-models-can-offer-greater-benefits.html
1Performance of Granite models conducted by IBM Research against leading open models across both academic and enterprise benchmarks - https://ibm.com/new/ibm-granite-3-0-open-state-of-the-art-enterprise-models