Back to Articles
Open-weight models such as NVIDIA Nemotron 3.5 Lightning allow cost-efficient, self-hosted inference for small and large businesses while also allowing for substantial customization. As a self-hosting software company with a track record of customizing LLMs, Nemotron 3.5 Lightning sparked our interest.
For context, TNG is a software consultancy with 930 employees in Munich. We provide LLM inference in Europe with currently more than 250 GPUs in service. Our inference endpoints serve more than 10 billion tokens per day. We host the latest open-weight models available, such as Kimi K3, GLM 5.2 and the latest Nemotron variants. Before using it, each new model is thoroughly tested through a comprehensive set of benchmarks that we put together from a list of industry best practice benchmarks. Our models are hosted in a highly secured cluster with zero data retention, a no-log policy and served from within Germany. We use hardware based Confidential Computing if extra security requirements are needed. Our research department has created models of their own based on model merges and post training of existing models, making the models more efficient than their parents in token usage or adding extra capabilities.
