SWE-rebench is a dataset and benchmark for code agents based on large models, developed by our AI R&D team. A curated subset of this dataset powers our leaderboard for Software Engineering LLMs, which we update every month with fresh tasks and model results. Our work is described in the arXiv paper, which on the release day became the #2 Paper of the Day on HF.

Scaling from lab to production: TrialHub and ELIXIR BioHackCloudScaling from lab to production: TrialHub and ELIXIR BioHackCloud

• Our customer TrialHub is a data intelligence platform making clinical trials efficient and patientcentric. Read the story about how Nebius assists TrialHub — the goal was to deliver quantifiable insights from unstructured, text-heavy data, scaling to production in days.

• BioHackCloud is advancing confidential computing and crossborder bioinformatics. Developed in the ELIXIR BioHackathon and enabled by Nebius, the project is designed as a reference implementation for a standards-based, federated multi-cloud platform.

This summer, our tech docs grew at a record paceThis summer, our tech docs grew at a record pace

• Networking in Kubernetes docs keep expanding — in Managed Kubernetes, you can now assign public IP addresses from a fixed set or convert a dynamically assigned public IP into a reusable allocation. We also added a maintenance overview with step-by-step instructions, a new guide on how to upgrade the version and create preemptible node groups.

• In Compute, networking guides got a boost too. Learn how to create virtual machines with static or dynamic IPs, create a VM without a public IP or get the IPv4 public range. We also covered attaching and detaching a filesystem from an existing VM.

• Object Storage deletion made easier. It can be hard to remove all objects from a bucket manually. We explained how to delete non-empty buckets after a delay.

• Observability agent gets a walkthrough. A new guide shows how to install, configure and update Nebius O11y Agent for metrics and logs ingestion.

• Managed PostgreSQL keeps growing. The service is now available in the us-central1 region, and the backups documentation is expanded with clear instructions for both automatic and manual backups.