Navan taps NVIDIA Nemotron 3.5 Super VL for AI travel agents under an open-model push

Navan, a publicly traded business travel and expense platform, said NVIDIA’s Nemotron 3.5 Super VL is now powering several of its AI travel agents. The announcement, published October 16, positions the company’s agent work inside an “open model strategy” that aims to deploy model capabilities in real-world enterprise workflows rather than limiting experimentation to a single vendor or approach.

In its update, Navan described the move as part of a broader effort to build and run AI agents for travel operations. While the company did not detail which specific travel tasks the agents perform, it said the new NVIDIA model is already in use, suggesting the deployment is at least at the stage where agents can support internal or customer-facing journeys.

NVIDIA, for its part, has been promoting model and platform options across AI development and enterprise use cases, including models intended to be adaptable for different application needs. Navan’s choice of Nemotron 3.5 Super VL ties its agent roadmap to a model family that NVIDIA is marketing for high performance in large language and multimodal-oriented scenarios, though the Navan announcement did not break out technical benchmarks.

The “open model strategy” framing matters because enterprise buyers often want flexibility around model providers, deployment environments, and integration paths. By emphasizing openness, Navan is implicitly arguing that AI agents for travel can be built to switch or expand model backbones over time, rather than treating the model choice as a fixed, one-time decision.

Market watchers will also note that business travel is a complex domain. It involves policy constraints, booking rules, change and cancellation flows, and expense documentation requirements, all of which can be difficult to encode with traditional workflow automation alone. AI agents can help by interpreting requests, routing them to the right actions, and translating travel intents into operational steps, but the quality of those steps typically depends heavily on the underlying model.

Still, the announcement leaves several questions unanswered. Navan did not disclose timelines beyond saying the model is now powering “several” agents, nor did it provide performance metrics, adoption numbers, or details on whether any user groups are involved. It also did not specify whether the deployment runs on NVIDIA infrastructure, whether it involves fine-tuning, or what safeguards and evaluation processes are in place for travel and expense outcomes.

As Navan expands its AI agent footprint, the next key items to watch are proof points. That includes published case studies, measured improvements in agent reliability or resolution speed, and clearer disclosure on how the company evaluates accuracy in travel booking and policy compliance. For NVIDIA, the partnership is another announcement that its models are finding traction in practical enterprise applications beyond pure research or internal prototypes.

For now, investors and customers are left with the headline-level confirmation that Navan has integrated Nemotron 3.5 Super VL into its AI agents. The real test will come as Navan reports more concrete results and as the company clarifies the scope of the deployment and the operational impact it is already delivering.