Selectel Adds New AI Models to Its AI Router, 4xH200 NVLink GPU Host and Other Product Updates

Selectel introduced several updates in July. The AI router now supports recently released SOTA models including Kimi K3, Qwen3.8 Max, DeepSeek V4 Flash, and image generation models such as the Flux.2 family from Black Forest Labs. The router provides access to leading LLMs through one account and a single API key, with token consumption paid in rubles on a pay-as-you-go basis. A new GPU host offers a physical machine with 4x NVIDIA H200 NVL PCIe (each with 141 GB HBM3e, totaling 564 GB) fully dedicated to one client's VM, with NVLink bridges enabling about 7x faster data exchange than PCIe. This configuration is suitable for inference and fine-tuning of large open-weight models, long contexts, and multimodal pipelines. Selectel also added audit log export to S3-compatible object storage, management of 1C technological journal in the control panel, Prometheus metric export from ready 1C cloud, expanded network disk geography to new availability zones, added a new dedicated server configuration DL-40 with AMD EPYC 7443P, simplified SELECTOS 2.0 installation, improved Managed Kubernetes with support for Kubernetes 1.36 and Open Kernel Module GPU drivers, added Ubuntu 26.04 LTS, new ML/AI applications (Hermes, RAGFlow, OpenClaw, Label Studio), moved application settings to the order form, implemented service dashboards for infrastructure monitoring, added a recursor in private DNS (available via Terraform and API for now), and improved DBaaS with simplified public IP connection and pg_uuidv7 extension for PostgreSQL.