FriendliAI Adds NVIDIA Nemotron 3 Super to AI Deployment Platform

According to a recent LinkedIn post from FriendliAI, the company is offering NVIDIA’s Nemotron 3 Super model on its platform from day one availability. The post highlights that developers can deploy this hybrid Mixture-of-Experts model through Dedicated Endpoints tailored for multi-agent and specialized agentic AI applications.

The LinkedIn post describes several technical attributes, including a hybrid Mamba–Transformer architecture aimed at higher token-generation efficiency and improved reasoning quality. It also notes claims of leading benchmark performance on GPQA Diamond, AIME 2025, LiveCodeBench, IFBench, and BFCL, as well as a “thinking budget” feature designed to better control inference costs.

Nemotron 3 Super is described as having 120B total parameters with 12B active and a context window of up to 1M tokens, positioning it toward large-scale, mission-critical workloads. The post further points to FriendliAI’s stated 99.99% uptime service-level objective as a foundation for running such workloads with enterprise-focused reliability.

For investors, the post suggests that FriendliAI is aligning closely with NVIDIA’s advanced model ecosystem and targeting high-end reasoning and agentic AI use cases. This alignment may enhance the company’s competitive position in AI infrastructure, potentially driving incremental demand from enterprise and developer customers seeking scalable, benchmark-competitive models.

By emphasizing cost-optimized reasoning and predictable inference, the content implies a focus on unit-economics-sensitive workloads, which could make the platform more attractive for production deployments. If adoption of Nemotron 3 Super on FriendliAI’s infrastructure scales, this could support higher recurring revenue and strengthen the firm’s role within the AI model deployment value chain.