IBM Granite-4.1-8b Advances Multilingual Chat And Tool Assistants

IBM has released Granite-4.1-8b, a new large language model designed for instruction following and chat tasks. This 8B parameter model is built to handle long contexts and complex requests. It functions as an optimized version of the previous Granite base model.

The Granite Team at IBM developed this release to improve how AI interacts with users and external tools. They used a combination of open-source datasets and custom synthetic data to train the model. Through rigorous fine-tuning and reinforcement learning, the developers enhanced its ability to follow specific directions and call functions.

Multilingual capabilities and tool integration

• Enhanced tool calling for connecting to external APIs.
• Support for 13 different languages including English, German, and Chinese.
• Strong performance in code-related tasks and text extraction.
• Advanced capabilities for retrieval augmented generation (RAG).
• Improved instruction following via supervised fine-tuning.

This model is useful for building specialized AI assistants that need to perform specific business logic. It also serves as a foundation for developers creating agents that must interact with software tools. People running local hardware can benefit from its efficient 8B parameter size.

Development and safety notes

The model is primarily trained on English data, so performance in other languages may vary. Users can improve accuracy in non-English tasks by providing a few examples in their prompts. IBM suggests that the community perform their own safety testing before using the model in professional environments.

While this model has been aligned by keeping safety in consideration, the model may in some cases produce inaccurate, biased, or unsafe responses to user prompts.