IBM's new enterprise AI models are more powerful than anything from OpenAI or Google

IBM is zooming along with new open-source Granite Large Language Models (LLM) releases every few months. Granite 3.1  is the latest generation model, building upon the success of Granite 3.0. The model offers enhanced capabilities and performance optimized for business applications.

Also: Gemini Advanced users can now access Google’s most experimental model

The family of Granite 3.1 models boasts an impressive 128K token context window, a substantial increase from their predecessors. This expansion allows the models to process and understand much larger amounts of text — equivalent to approximately 85,000 English words — enabling more comprehensive analysis and generation tasks. By comparison, OpenAI’s ChatGPT 3, which ignited the AI revolution, could handle only 2,000 tokens.

Outperforming the competition

Big Blue claims its new Granite 8B Instruct model outperforms its rivals, such as Google Gemma 2, Meta Llama 3.1, and Qwen 2.5, on HuggingFace‘s OpenLLM Leaderboard benchmarks.

Also: Want generative AI LLMs integrated with your business data? You need RAG

The Granite 3.1 family includes dense models and Mixture of Experts (MoE) variants. IBM states its Granite 2B and 8B models are text-only dense LLMs trained on over 12 trillion data tokens. The dense models are designed to support tool-based use cases and for retrieval augmented generation (RAG), streamlining code generation, translation, and bug fixing.

The MoE models are trained on over 10 trillion tokens of data. IBM claims these models are ideal for deployment in on-device applications with low latency.

Improved graphics and language capabilities

The new release also introduces image-in/text-out functionality, broadening the models’ applicability for businesses working with graphics.

Granite 3.1 also offers improved foreign language proficiency. Alongside English, it can now work with a dozen languages, including German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Simplified Chinese. In short, if you have an international business, IBM has an LLM you can use.

As before, the new Granite LLMs are licensed under the Apache 2 open-source license. IBM also maintains its commitment to openness by disclosing the training datasets and providing comprehensive documentation on model development and performance.

Also: Generative AI brings new risks to everyone. Here’s how you can stay safe

If you’re worried about intellectual property issues and AI, IBM offers an uncapped indemnity for third-party IP claims related to the use of its models. You may not be concerned about such potential legal problems yet, but I assure you that you will be.

Powerful, trustworthy AI for enterprises

Granite 3.1 models are available on IBM’s Watsonx platform; cloud service providers like Google Vertex AI; and AI platforms including Hugging Face, NVIDIA (as NIM microservices), Ollama, and Replicate.

The release of Granite 3.1 is poised to accelerate AI adoption in enterprise settings. By combining these models with proprietary data using techniques like IBM’s InstructLab, businesses can potentially achieve task-specific performance rivaling larger models at a fraction of the cost.

As the AI landscape evolves rapidly, IBM’s Granite 3.1 represents a significant step forward in providing enterprises with powerful, efficient, and trustworthy AI tools to drive innovation and solve complex business challenges.