IBM ships Granite 4.1 8B New — small-model enterprise-deployable LLM with Apache 2.0 license joins the open-weight production cohort
IBM released Granite 4.1 8B New in May 2026, the latest in the Granite family of enterprise-focused open-weight LLMs. At 8 billion parameters under Apache 2.0 license, the model targets the deploy-on-your-own-infrastructure enterprise cohort that needs a credible small-model option without the compliance constraints of Chinese open-weight alternatives.
The 8B parameter size is a deliberate choice. IBM's Granite line targets the corporate deployment use case where the model has to run on the customer's own infrastructure — typically Intel Xeon CPUs, AMD EPYC, or modest GPU configurations rather than NVIDIA H100 clusters. At 8B parameters with 4-bit quantization, Granite 4.1 fits in roughly 4-6GB of RAM and runs at usable inference speeds on commodity hardware. That's the form factor that makes it viable for the broad-deploy-everywhere use cases that don't justify the infrastructure cost of larger models.
The Apache 2.0 license is the procurement-relevant part. For US-headquartered enterprise customers, the procurement-team analysis goes: Chinese open-weight options (Qwen, DeepSeek, GLM, Kimi) face vendor-due-diligence questions about origin-country risk; European open-weight (Mistral) is procurement-friendly but limited to a smaller product line; US-headquartered open-weight is Mistral... and IBM Granite. The IBM brand carries enterprise procurement trust that startups can't match. For risk-averse Fortune 500 procurement, Granite is the path of least resistance into open-weight LLM deployment.
Hugging Face — Best Open-Source LLM Models 2026 → · Presenc AI — Open-Source LLM Landscape 2026 15+ Model Families Mapped → · LLM Stats — AI Updates Today May 2026 →