Exciting news for tech enthusiasts and businesses alike: IBM has just unveiled the latest addition to its Granite family—Granite 3.2! This new model comes with fresh enhancements and impressive upgrades that are sure to capture attention. For those unfamiliar with IBM Granite, it is a suite of open, high-performance AI models specifically designed for businesses and optimized for scaling AI applications. What sets Granite apart is its commitment to open-source accessibility, making these advanced tools available to everyone.
The Granite 3.2 particularly the Granite 3.2 8B instruct version, introduces groundbreaking experimental chain-of-thought reasoning capabilities. This feature significantly enhances the model’s ability to comprehend and execute complex instructions while maintaining exceptional overall performance. As the latest development in the third generation of Granite models, Granite 3.2 marks a pivotal advancement in the series, moving beyond basic language processing.
It showcases promising new features, including IBM’s first official vision language model (VLM), alongside numerous improvements to the efficiency and versatility of existing offerings. IBM continues to prioritize the creation of practical, enterprise-ready models that deliver state-of-the-art performance with increasingly optimized parameters.
Here are some benchmark scores for
| Benchmark | Granite |
| ArenaHard | 55.23 |
| Alpaca-Eval-2 | 61.16 |
| IFEval | 73.57 |
| MMLU | 66.93 |
| PopQA | 28.08 |
| TruthfulQA | 66.37 |
In summary, IBM is relentlessly pursuing innovative advancements in AI, with the new Granite 3.2 model paving the way for developers to efficiently enhance their businesses and projects. You can find this exciting new model available on Hugging Face, consistent with IBM’s commitment to open-source development.