Eighteen months ago, selecting an AI model resembled making a conventional architecture decision: Evaluate the options, choose one, and build on it. That approach was reasonable then, but it’s now as outdated as an 18-month-old AI model.
Today, model capability advances on cycles measured in weeks. Reasoning quality has improved substantially. Context windows have expanded by orders of magnitude. Open-weight models have closed a divide that most of the industry expected would remain open far longer.
For enterprises, the implication is clear: Commit to a single model, no matter how capable it is today, and you’ve transferred the risk of a fast-moving research field directly to your business. Every time the frontier moves—because it will move—your organization absorbs the disruption.
At ServiceNow, our AI models strategy is built around a different premise: We think about the model layer as a portfolio that we manage on our customers’ behalf. It spans frontier and open-weight models with choice and flexibility grounded in rigorous evaluation. Three principles guide our approach.
Enterprise AI workloads vary widely. Classifying a high volume of incoming requests, summarizing a long case history, and coordinating a multistep agentic task across several systems place fundamentally different demands on a model. Optimizing a single benchmark score obscures that difference. Instead, we optimize for the work to be done.
We evaluate models against the workflows our customers actually run, measuring accuracy, latency, and end-to-end performance. That evaluation discipline, not a preference for any one provider, determines what runs where. The result is a portfolio in which each model is deployed against the work it handles best.
Importantly, via ServiceNow AI Control Tower, customers can always override any model provider defaults to align with their corporate policies.
For the most demanding agentic work—reasoning across systems, planning multiple steps ahead, or recovering when an initial approach within an agentic workflow fails—frontier models represent the strongest capability available. We partner and integrate with OpenAI, Anthropic, Google, and others.
Customers who already hold a commercial relationship with a provider continue operating under their existing terms or bring their own key, connected directly through the ServiceNow AI Platform.
We’re model agnostic and cloud agnostic, meaning our customers have the choice to use what’s best for their business. And they retain the ability to enable or disable any provider via ServiceNow AI Control Tower.
Open-weight models have earned a central place in the enterprise. We make them available to every customer.
Open weights give you visibility and control over where the model runs. For any organization operating under data residency requirements, sector-specific regulation, or strict internal governance, that visibility changes what’s possible.
ServiceNow has a long history of hosting highly performant open-weight models, including Now LLM, StarCoder, Apriel, and GPT-OSS. That commitment holds. We’ll continue to offer a suite of ServiceNow-hosted open-weight models for all customers.
The AI model ecosystem is rapidly evolving with dramatic performance improvements and new capabilities. We’re dedicated to a continuous, rigorous approach: As new models and architectures emerge, we evaluate them against real-life enterprise workloads and embrace those that deliver clear benefits.
This approach guides the model portfolio: open-weight models that offer transparency and control, and frontier models that provide advanced capabilities. We’re committed to adapting and evolving our strategy as the ecosystem develops and our customers’ needs evolve, helping to ensure we always deliver the best possible solutions.
The core objective is simple: Customers should benefit from advances in AI without absorbing the cost of each transition. That means no re-architecture, no migration project, and no disruption to work in progress. The model layer changes. The workflows keep running.
Find out how ServiceNow can help you discover, secure, and measure any AI.
