AI inference platform company FriendliAI announced that it is participating as an official launch partner of NVIDIA and will provide NVIDIA’s new open AI model “Nemotron 3” through its own platform.

Through this collaboration, enterprises and developers can easily deploy and operate the Nemotron 3 Nano model on the FriendliAI platform without building separate, complex infrastructure. FriendliAI emphasizes fast response speed and cost efficiency as its strengths and provides an AI operating environment suitable for real-world service environments. On FriendliAI’s high-performance inference environment, the Nemotron 3 model can be used to operate AI agents in a stable manner.

Jun Byung-Gon, CEO of FriendliAI, said, “The combination of Nemotron 3 and the FriendliAI platform is an important milestone that enables the implementation of ‘agentic AI,’ in which AI makes its own decisions and performs tasks, at more realistic cost and performance levels,” adding, “Our goal is to provide enterprises with both the performance and efficiency required to operate large-scale AI services.”

The Nemotron 3 model is an AI model specialized for tasks that require complex reasoning and logical judgment, and it features a Mixture-of-Experts (MoE) architecture based on a hybrid Mamba-Transformer. The Nano model is designed with a 1-million-token context window and can be used across various industries, including software development, finance, retail, and cybersecurity. In particular, it adopts a structure that uses computing resources efficiently, enabling reductions in operating costs while maintaining performance.

When running the Nemotron 3 Nano model on the FriendliAI platform, it delivers stable performance through optimized GPU utilization, MoE serving optimization technology, and autoscaling functionality. When using dedicated endpoints, GPU costs can be reduced by more than 50% compared with conventional methods. In addition, it supports an API compatible with OpenAI, making integration with existing AI services straightforward.

CEO Jun said, “For companies considering AI adoption, the biggest concerns are cost and operational complexity,” and added, “FriendliAI will continue to provide infrastructure that enables easier and more rational utilization of high-performance AI.”

Meanwhile, FriendliAI supports more than 480,000 AI models through dedicated AI inference, serverless APIs, and container-based solutions, and it supports the operation of AI inference services in the production environments of a broad range of customers, from startups to large enterprises.

Popular News