● The newly launched Granite 3.0 8B and 2B models are released under the permissive Apache 2.0 license and have demonstrated strong performance in many academic and enterprise benchmark tests, capable of surpassing or matching models of similar scale.

● The newly launched Granite Guardian 3.0 model provides IBM's most comprehensive guardrail function to promote safe and trustworthy artificial intelligence.

● The newly launched Granite 3.0 expert mixture-of-experts models enable extremely efficient inference and low latency, suitable for CPU-based deployments and edge computing.

● The brand-new Granite time series model achieves state-of-the-art performance in zero-shot/few-shot prediction, surpassing models that are ten times larger.

● IBM has launched the next generation of Granite-based watsonx Code Assistant for general coding; new tools for building and deploying AI applications and agents are debuted in watsonx.ai.

● It is announced that Granite will become the default model for Consulting Advantage, an AI-driven delivery platform used by IBM's 160,000 consultants to provide new solutions to customers faster.

At IBM's annual TechXchange conference held on October 21, US time, IBM announced the launch of its most advanced AI model family to date - Granite 3.0. IBM's third-generation Granite flagship language model can surpass or match the models of similar leading model providers in many academic and industry benchmark tests, demonstrating powerful performance, transparency, and security.

In line with the company's commitment to open source AI, Granite models are released under the permissive Apache 2.0 license. With their unique combination of performance, flexibility, and autonomy, they can serve enterprise customers and the entire community.

IBM's Granite 3.0 family series includes:

● General/Language models: Granite 3.0 8B Instruct, Granite 3.0 2B Instruct, Granite 3.0 8B Base, Granite 3.0 2B Base

● Guardrail and security models: Granite Guardian 3.0 8B, Granite Guardian 3.0 2B

● Expert mixture-of-experts models: Granite 3.0 3B-A800M Instruct, Granite 3.0 1B-A400M Instruct, Granite 3.0 3B-A800M Base, Granite 3.0 1B-A400M Base

The new Granite 3.0 8B and 2B language models are designed as the "main force" models for enterprise-level AI, capable of providing strong performance in tasks such as retrieval-enhanced generation (RAG), classification, summarization, entity extraction, and tool usage. These compact and versatile models are intended to be fine-tuned with enterprise data and seamlessly integrated into various business environments or workflows.

Many large language models (LLMs) are trained on publicly available data, while the vast majority of enterprise data remains unused. By combining small Granite models with enterprise data, especially using the revolutionary alignment technology InstructLab launched by IBM and RedHat in May, IBM believes that enterprises can achieve task-specific performance comparable to large models at only a fraction of the cost (based on the observed cost range of 3 to 23 times lower than large cutting-edge models in several early proof-of-concepts).

The release of Granite 3.0 reaffirms IBM's commitment to building transparency, safety, and trust in AI products. The Granite 3.0 technical report and responsible use guide provide a description of the datasets used to train these models, detail the applied filtering, cleaning, and processing steps, and comprehensively demonstrate the performance results of the models in major academic and enterprise benchmark tests.

Crucially, IBM provides intellectual property indemnification for all Granite models on watsonx.ai, aiming to enhance the confidence of enterprise customers in integrating enterprise data into the models.

Raising the bar: Granite 3.0 benchmark tests

The Granite 3.0 language models also show good results in raw performance.

In the standard academic benchmark tests defined by Hugging Face's OpenLLM leaderboard, the overall performance of the Granite 3.0 8B Instruct model on average leads the state-of-the-art performance of similar-sized open source models from Meta and Mistral. In IBM's advanced AttaQ security benchmark test, the Granite 3.0 8B Instruct model leads Meta and Mistral's models in all measured security dimensions.

In core enterprise tasks such as retrieval-enhanced generation (RAG), tool usage, and cybersecurity, the average performance of the Granite 3.0 8B Instruct model is better than that of open source models of similar scale from Mistral and Meta.

The Granite 3.0 models are trained on over 12 trillion tokens from 12 different natural languages and 116 different programming languages, using a novel two-stage training method that leverages thousands of experimental results aimed at optimizing data quality, data selection, and training parameters. By the end of this year, the 3.0 8B and 2B language models are expected to support an expanded 128K context window and multimodal document understanding capabilities
.

IBM has demonstrated the perfect balance of performance and inference cost by providing its Granite expert mixture-of-experts (MoE) institutional models, Granite 3.0 1B-A400M and Granite 3.0 3B-A800M. These smaller and lightweight models can be used for low-latency applications and CPU-based deployments.

IBM also announced an updated version of its pre-trained Granite time series model, with the earlier version released earlier this year. These new models are trained on three times more data and perform well in all three major time series benchmark tests, surpassing the performance of models from companies like Google and Alibaba that are ten times larger. The updated models also provide greater modeling flexibility and support for external variables and rolling predictions.

Granite Guardian 3.0: Ushering in a new era of responsible AI

As part of this release, IBM also launched a new Granite Guardian model series that allows application developers to implement safety guardrails by checking user prompts and LLM responses to detect various risks. The Granite Guardian 3.0, 8B, and 2B models offer the most comprehensive risk and hazard detection functions on the market today.

In addition to hazard dimensions such as social bias, hatred, toxicity, blasphemy, violence, and jailbreaks, these models also provide a series of unique retrieval-enhanced generation (RAG)-specific checks such as factuality, context relevance, and answer relevance. In extensive testing against 19 safety and RAG benchmarks, the Granite Guardian 3.0 8B model has overall higher accuracy in hazard detection on average than all three generations of Meta's Llama Guard models. Its overall performance in hallucination detection is also comparable to specialized hallucination detection models WeCheck and MiniCheck.

Although the Granite Guardian models are derived from the corresponding Granite language models, they can be used with any open or proprietary AI model to implement safety safeguards.

Availability of Granite 3.0 models

The entire Granite 3.0 model suite and updated time series models can be downloaded under the permissive Apache 2.0 license on HuggingFace. The instruction variants of the new Granite 3.0 8B and 2B language models and the Granite Guardian 3.0 8B and 2B models are now available for commercial use on IBM's watsonx platform. Some Granite 3.0 models will also be provided as NVIDIA NIM microservices and through Google Cloud's Vertex AI Model Garden integration with HuggingFace.

To provide developers with multiple options and ease of use and support local and edge deployments, a selection of Granite 3.0 models is also available on Ollama and Replicate. The latest generation of Granite models expands IBM's powerful catalog of open source LLMs. IBM partners with ecosystem partners such as AWS, Docker, Domo, Qualcomm Technologies, Inc. (through its Qualcomm® AI Hub), Salesforce, and SAP to integrate multiple Granite models into these partners' products or offer Granite models on their platforms, providing greater choice for global enterprises.

From assistants to agents: Realizing the future of enterprise-level AI

IBM is advancing enterprise-level AI through a series of technologies, from models and assistants to the tools needed to tune and deploy AI for a company's unique data and use cases. IBM is also paving the way for future AI agents that can be self-directed, reflective, and perform complex tasks in a dynamic business environment.

IBM is continuously evolving its portfolio of AI assistant technologies, from watsonx Orchestrate that helps companies build their own assistants through low-code tools and automation, to various pre-built assistants for specific tasks and domains such as customer service, human resources, sales, and marketing. Organizations around the world have already used watsonx Assistant to help them build AI assistants for tasks such as answering daily questions from customers or employees, modernizing mainframes and legacy IT applications, helping students explore potential career paths, or providing digital mortgage support for homebuyers.

IBM also announced the upcoming release of the next generation of watsonx Code Assistant, which is powered by Granite code models and can provide general coding assistance for languages such as C, C++, Go, Java, and Python, as well as advanced application modernization capabilities for enterprise-level Java applications. Granite's code capabilities can now also be accessed through the IBM Granite.Code Visual Studio Code extension.

IBM also plans to release new tools to help developers build, customize, and deploy AI more efficiently using watsonx.ai, including agent frameworks, integrations with existing environments, and low-code automation for common use cases such as RAG and agents.

IBM is focused on developing AI agent technologies with higher autonomy, complex reasoning capabilities, and multi-step problem-solving abilities. The initial version of the Granite 3.0 8B model supports key agent functions such as advanced reasoning and highly structured chat templates and prompt styles for implementing tool usage workflows. IBM also plans to introduce new AI agent chat capabilities in IBM watsonx Orchestrate, leveraging agent capabilities to coordinate AI assistants, skills, and automation to help users improve overall team productivity [viii]. IBM plans to continue building agent capabilities into its product portfolio in 2025, including pre-built agents for specific domains and use cases.

Expanding the AI-driven delivery platform to enhance IBM consultants with AI

IBM also announced a significant expansion of its AI-driven delivery platform, IBM Consulting Advantage. This multi-model platform contains AI agents, applications, and methods (such as reusable frameworks), empowering 160,000 IBM consultants to provide value to customers faster and better at lower cost.

As part of the expansion, the Granite 3.0 language model will become the default model for Consulting Advantage. With the performance and efficiency of Granite, IBM Consulting will be able to help maximize the return on investment in generative AI projects for IBM customers.

Another key part of the expansion is the introduction of IBM Consulting Advantage for Cloud Transformation and Management and IBM Consulting Advantage for Business Operations. Each includes domain-specific AI agents, applications, and methods that incorporate IBM's best practices, enabling IBM consultants to help customers accelerate cloud and AI transformation tasks (such as code modernization and quality engineering) or implement transformations and operations across domains (such as finance, human resources, and procurement).

About IBM

IBM is a global leader in hybrid cloud, artificial intelligence, and enterprise services, helping customers in more than 175 countries and regions gain business insights from their data, simplify business processes, reduce costs, and gain a competitive edge in the industry. More than 4,000 government and enterprise entities in critical infrastructure sectors such as financial services, telecommunications, and healthcare rely on IBM's hybrid cloud platform and Red Hat OpenShift to achieve digital transformation quickly, efficiently, and securely. IBM's breakthrough innovations in artificial intelligence, quantum computing, industry cloud solutions, and enterprise services provide our customers with open and flexible choices. The long-term commitment to corporate integrity, transparent governance, social responsibility, inclusive culture, and service spirit is the cornerstone of IBM's business development.