IBM AI Updates Include Granite 3.0, Watsonx Upgrades

An expansion to IBM Consulting Advantage and greater model access across partner platforms were also part of the updates.

IBM has revealed its Granite 3.0 flagship language models, new capabilities in the Watsonx artificial intelligence platform, an expansion to its Consulting Advantage AI-powered delivery platform and greater model access across partner platforms.

The news came during IBM’s annual TechXchange event, which runs Monday to Thursday in Las Vegas. As part of the news, IBM also announced an update to its pre-trained Granite models.

Rob Thomas, IBM’s senior vice president of software and chief commercial officer, told CRN during a virtual press event ahead of the Granite 3.0 reveal that system integrators and other partners have seen opportunities selling customers on the Granite models plus opportunities in governing models as customers scale AI use.

Thomas also pointed to the Armonk, N.Y.-based vendor’s model indemnification policy as a unique differentiator for IBM as the AI vendor of choice for solution providers.

“You think about the confidence that gives an MSP, or a system integration partner, that IBM is standing behind everything we've done from a training perspective, that's pretty dramatic,” Thomas said.

IBM AI Updates

During the virtual press event, Ritika Gunnar, IBM’s general manager of data and AI, told CRN that the vendor’s solution providers can leverage its smaller language models and the InstructLab open source model enhancement project co-launched with IBM subsidiary Red Hat to customize models with their data and create unique domains.

“Because the base model itself has transparency and not only the data, but the rights that you have with that foundation model, and then you can apply your own domain-specific data to that, it becomes very lucrative, not only from a cost perspective, but from a transparency perspective and for a governance perspective for organizations to build with,” Gunnar said.

On the call, Dario Gil, IBM senior vice president and director of research, said that the licensing strategy IBM is pursuing for AI allows users to leverage their own brand.

“Very often, bespoke licenses prevent you from attaching a brand on top of a base capability, Gil said. “We impose no such thing. They can take that. They can create their own brand. They can express it and however they see fit and grow with it.”

The vendor has pledged more AI agent capabilities across its portfolio in 2025, which will include pre-built agents for specific domains and use cases–adding to the popularity of agent AI offerings coming from rivals such as Microsoft, Google and Salesforce.

In a blog post Monday, Kate Woolley–IBM’s channel chief whose formal title is general manager of the IBM ecosystem–said that the vendor’s tens of thousands of solution providers help end customers build trust in AI offerings, scale them and manage costs and complexity.

“The IBM Ecosystem acts as the linchpin to ensure clients can access and put AI advances like these to work for their business,” Woolley said. “Comprised of ISVs, hyperscalers, resellers and distributors, service providers, technology partners, consultancies, systems integrators and MSPs, these partners are bringing our state-of-the-art technology to business users at scale.”

IBM Granite 3.0 AI Models

IBM bills Granite 3.0 as its most advanced family of AI models to date, according to a statement Monday. Part of the 3.0 family are new Granite 8B and 2B “workhorse” models used for retrieval augmented generation (RAG), classification, summarization, entity extraction and tools use. Users can fine-tune the new 8B and 2B models with enterprise data and integrate them into any business environment or workflow, according to IBM.

By the end of the year, the 8B and 2B models will support extended 128K context length and multi-modal document understanding capabilities, according to IBM.

The new models will be embedded in the December release of Red Hat Enterprise Linux (RHEL) AI and OpenShift AI, Gil said during the press event.n

The initial Granite 8B model release supports agentic AI capabilities including advanced reasoning and a chat template and prompting style for tool use workflows.

Also part of the family are Granite mixture of experts (MoE) architecture models Granite 1B A400M and 3B A800M. Users can leverage these smaller, lightweight models for low-latency applications and central processing unit (CPU)-based deployments.

The new family of Granite Guardian guardrail and safety models allow app developers to check user prompts and model responses for a variety of risks. They are derived from corresponding Granite language models, but they can implement guardrails alongside any open or proprietary AI model, according to the vendor.

The models have RAG-specific checks for groundedness, context relevance and answer relevance while also accounting for social bias, hate, toxicity, profanity, violence and jailbreaking, among other dimensions, according to IBM.

Granite 3.0 models are under a fully permissive Apache 2.0 license, aligning with the vendor’s open-source AI commitment, according to IBM. The license differentiates the models from competing ones by giving enterprises and the community more rights along with the models’ improved performance.

Availability On Watsonx, Partner Platforms

Now available for commercial use on IBM’s Watsonx artificial intelligence platform are instruct variants of the new Granite 8B and 2B language models and the 8B and 2B Granite Guardian models.

IBM plans to make select Granite 3.0 models available on Nvidia’s NIM stack, Google Vertex, Domo and other partner platforms, the vendor said Monday.

A curated set of Granite language and mixture-of-experts (MoE) models are available on Ollama and Replicate.ai. Users can download on Hugging Face under the Apache 2.0 license the whole suite of 3.0 models and updated time series models, according to IBM.

Other partnerships IBM highlighted in a separate statement Monday include Amazon Web Services (AWS) now offering several Granite models for its Sagemaker JumpStart machine learning (ML) hub through its AWS Marketplace.

Users of the Amazon Bedrock generative AI apps builder can also access IBM Granite models, according to IBM.

Qualcomm, Salesforce and SAP have made Granite models available to their developer communities and ecosystems, according to IBM. Samsung SDS, a consulting and outsourcing subsidiary of Samsung, is piloting Granite time series models for anomaly detection and open-sourced Granite models through Watsonx for supply chain relationship management use cases.

Watsonx Updates

IBM’s new generation of Watsonx Code Assistant brings general-purpose coding assistance for C, C++, Go, Java, Python and other languages. The tool also helps with modernization capabilities for Enterprise Java apps.

The vendor also said to expect agentic frameworks, existing environment integrations, low-code automations for RAG, agents and other common use cases and other new developer tools in Watsonx.ai.

As for Watsonx Orchestrate, IBM has a new AI agent chat feature that will allow users to orchestrate AI assistants, skills and automations, according to the vendor.

IBM Consulting Advantage

Among IBM’s many AI updates was the expansion of its Consulting Advantage AI-powered delivery platform, which has adopted Granite 3.0 as its default language model.

IBM Consulting Advantage has added a Cloud Transformation and Management offering and a Business Operations offering, each with domain-specific AI agents, applications and methods to go along with IBM intellectual property and best practices.

The two new Consulting Advantage offerings aim to help speed up code modernization and other tasks and help operations across finance, procurement and other domains, according to IBM.

The Consulting Advantage platform is used by 160,000 IBM consultants, according to the vendor. IBM Consulting is No. 6 on CRN’s 2024 Solution Provider 500

Mohamad Ali, senior vice president and head of IBM Consulting, told CRN during the virtual press event that Granite models are “important to our economic model going forward, as well as our flexibility in the types of projects we can do.”

“Starting with the license–the ability to then do derivative projects and create new intellectual property either for us consulting or for our clients,” Ali said. “We’re working with a bank right now, the bank wants to keep all the IP to themselves, understandably. … We have a hospitality client, and he said to me, ‘This thing's too expensive.’ And so he had to reserve something called prompt caching so he's not hitting the LLM all the time, because they're too expensive. Granite changes that.”