Navan Says Several Of Its AI Travel Agents In Production Are Being Powered By Nvidia Nemotron 3.5 Super VL

Just three weeks after receiving early access and beginning its work with NVIDIA, Navan trained and deployed the model. The deployment is the latest example of Navan’s open-model strategy: evaluating models for specific travel tasks, adapting them with Navan’s travel expertise, and self-hosting them to maintain control over performance, cost, and reliability.

Before deployment, Navan internally evaluated Nemotron 3.5 Super VL against its self-hosted production baseline and other models using consistent prompts, tools, travel scenarios and criteria. After fine-tuning with NVIDIA NeMo AutoModel VLM LoRA training recipe, Nemotron 3.5 Super VL delivered the strongest overall performance, beating or tying every competing model on six of seven measures, including having the lowest latency. The model is now in production and powering customer conversations at a lower cost while delivering greater value to Navan users.

Navan combines open and closed models to build AI travel agents that understand what travelers want, follow policies, and use the right tools to get the job done.