IBM has released Granite 4.0 Nano, which are tiny versions of its large language models. The objective is to enable AI to operate not in data centers, but directly on the edge and on your devices. Essentially, IBM is stating that giant models with hundreds of billions of parameters are unnecessary for many tasks. These smaller models, available with 1.5 billion and even 350 million parameters, have been trained on over 15 trillion tokens using novel methodologies. Crucially, they are available under the Apache 2.0 license and are compatible with popular environments such as vLLM and llama.cpp. This represents a direct challenge to resource-intensive models. For businesses, this means an end to dependence on expensive cloud infrastructure. Running AI locally promises to reduce infrastructure costs and make integration significantly more accessible. Expect new intelligent mobile applications and edge solutions that will not rely on the cloud but become integral parts of your devices. This move paves the way for more affordable and private AI, accessible even to those not yet prepared to invest millions in GPUs.

AI ToolsAI in BusinessOpen Source AIOn-Device AIIBM