ServiceNow’s Apriel 2.0 promises smarter AI with less hardware — but offers no benchmarks

ServiceNow has announced an updated version of its Apriel open-source language model. The model is built on Nvidia’s Nemotron framework and targets financial services, healthcare, and telecommunications industries, where regulators have made IT leaders cautious about proprietary AI systems, the company said in a statement.

ServiceNow said Apriel 2.0 delivers reasoning comparable to much larger models while running on less hardware. It handles multimodal inputs — screenshots, forms, diagrams — which ServiceNow said enterprises need for autonomous agent deployments.

The company, however, provided no benchmarks, inference cost comparisons, or data showing how the model stacks up against alternatives already in production.

The company announced Apriel 2.0 at Nvidia’s GTC conference in Washington, D.C., saying it would reach production by Q1 2026.

Smaller models, crowded field

ServiceNow built Apriel 2.0 on its earlier Apriel Nemotron 15B model, which it introduced at its Knowledge 2025 conference earlier this year and is now in production. The company also recently shipped Apriel-1.5-15B-Thinker, an open-source model it said runs on a single GPU.

The timing works well for ServiceNow’s strategy. The company cited a Gartner forecast predicting that by 2027, smaller, context-specific models will see usage volumes at least three times greater than general-purpose large language models. ServiceNow’s announcement included no performance benchmarks, no inference cost comparisons, and no data on how Apriel 2.0 stacks up against alternatives already in production.

Apriel 2.0 enters a crowded market for smaller open-source models targeting enterprises. Microsoft’s Phi-4, released in December 2024, runs 16 billion parameters and performs well on reasoning tasks despite its compact size. Meta’s Llama 3.3-70B handles dialogue in eight languages with a 128,000-token context window. Google’s Gemma 2 comes in configurations from 2 billion to 27 billion parameters, optimized for resource-constrained environments. Mistral’s models range from 1.3 billion to 13 billion parameters, emphasizing efficiency and integration speed.

These models all address the same enterprise concerns: reducing cloud computing costs, meeting data sovereignty requirements, and providing audit trails for regulated industries. What remains unclear is how Apriel 2.0 differentiates itself beyond its ServiceNow workflow integration.

“Open models give enterprises the transparency and control they need to specialize AI to their data, workflows, and trust standards,” Kari Briski, Nvidia’s vice president of generative AI for enterprise, added in the statement.

Government cloud integration

ServiceNow also announced it would integrate with Nvidia’s AI Factory for Government reference design, embedding its workflow software into Nvidia’s infrastructure blueprints. The integration carries the same Q1 2026 timeline.

ServiceNow outlined specific use cases: AI agents to handle gift card replacements and point-of-sale failures for retailers, citizen service request management for government agencies, and data center asset tracking for AI Factory operators.

The AI Factory for Government integration targets agencies facing pressure to modernize IT systems while regulators work out AI governance frameworks. ServiceNow stated that its workflows would comply with FedRAMP and NIST 800-53 requirements; however, federal procurement and security authorization processes typically extend well beyond product availability dates.

“The next wave of AI is about more than innovation. It’s about execution — how fast and how responsibly enterprises can put advanced intelligence to work,” Pat Casey, ServiceNow’s CTO and executive vice president for DevOps, said in the announcement.

ServiceNow provided no metrics on inference costs, no latency benchmarks, and no comparisons showing how Apriel 2.0’s resource requirements compare to existing enterprise deployments.

The Q1 2026 timeline puts both products months away. ServiceNow’s forward-looking statements acknowledged execution delays, regulatory changes, and uncertain market demand as risks.