Skip to main content
Nebius AI Cloud offers NVIDIA NIM microservices in Standalone Applications. The microservices are available in the eu-north1 and us-central1 regions.

​
About the microservices

NVIDIA NIM offers containers for self-hosting GPU-accelerated inference microservices, supporting both pretrained and custom AI models across cloud platforms, data centers, and RTX™ AI PCs and workstations. These NIM microservices provide industry-standard APIs for seamless integration with AI applications, development frameworks and workflows, while optimizing response latency and throughput for every foundation model and GPU pairing. Deploying NIM microservices in Nebius AI Cloud lets you use inference endpoints without creating and providing a personal API key for NVIDIA NGC, the cloud platform from NVIDIA. Nebius AI Cloud offers models from the NVIDIA BioNeMo Framework for computational drug discovery as NIM microservices:

​
Boltz-2 on NVIDIA NIM

Boltz-2 is an advanced foundation model for structural biology, delivering high accuracy in both structural prediction and affinity estimation. It is the first deep learning system to reach performance levels comparable to free energy perturbation (FEP) methods for predicting protein–small molecule binding affinities. It is showing strong benchmark correlations while being almost 1000× faster computationally. For more details, see the application page in the web console.

​
Evo 2 on NVIDIA NIM

Evo 2 is a biological foundation model that can process long genomic sequences while remaining sensitive to single-nucleotide variations. With 40 billion parameters, it captures the genetic code across all domains of life and stands as the largest AI model for biology to date. Its training was based on a dataset containing nearly 9 trillion nucleotides. For more details, see the application page in the web console.

​
GenMol on NVIDIA NIM

GenMol is a masked diffusion model trained on molecular Sequential Attachment-based Fragment Embedding (SAFE) representations for fragment-based molecule generation. It can function as a generalist model for a wide range of drug discovery tasks. These tasks include de novo molecule generation, linker design, motif extension, scaffold decoration and morphing, hit generation and lead optimization. For more details, see the application page in the web console.

​
MolMIM on NVIDIA NIM

MolMIM is a latent variable model developed by NVIDIA and trained unsupervised on a large dataset of molecules in SMILES format. It uses transformer architecture with Mutual Information Machine (MIM) learning to create an informative, clustered latent space. The model can generate novel molecules by sampling around a seed structure in this latent space. MolMIM also performs optimization with the CMA-ES algorithm to produce molecules with improved values of a desired scoring function. For more details, see the application page in the web console.

​
Pricing

In addition to computing and storage resources that your NVIDIA NIM microservice uses, Nebius AI Cloud charges you for the NVIDIA AI Enterprise License on an hourly basis, depending on the microservice’s number of GPUs. For more details, see Standalone Applications pricing in Nebius AI Cloud.

​
Getting started

To use an NVIDIA NIM microservice, deploy it. After that, you can use it to perform inference.