NVIDIA expanded its portfolio of open models designed for customization and local deployment across various industries. The new lineup includes language and vision models Nemotron 3.5 Super VL and Nemotron 3.5 Bolt, developed alongside European partners like Mistral and Black Forest Labs. NVIDIA also introduced the Kumo family of open-weight foundation models and libraries, aimed at forecasting and risk scoring from structured business data.

In addition to language and data models, NVIDIA unveiled Cosmos 3.5 world models for robotics and autonomous vehicles, Alpamayo 2 Edge for autonomous driving, BioNeMo for healthcare, and Ising for quantum computing. Enterprise adopters and post-training partners including SAP, Salesforce, Palantir, and Snowflake are integrating these open foundation models into domain-specific workflows.

Why it matters

  • Offers founders open-weight models optimized for tabular data predictions and physical AI simulation without vendor lock-in.

  • Allows operators to lower inference costs by fine-tuning domain-specific agents on proprietary infrastructure.

Source: blogs.nvidia.com