NVIDIA Expands Open AI Portfolio With Four New Models
The latest models for robotics, finance, and biology prioritize local data control and infrastructure integration.
Updated on Oct. 21, 2026 in Artificial Intelligence
Live Poll
Should businesses trust third-party artificial intelligence models for making core predictions and operational decisions?
NVIDIA has expanded its portfolio of open-source AI models, releasing new tools for structured business data, document processing, robotics, and protein sequence generation. The company also announced the upcoming availability of its Alpamayo 2 Edge model for autonomous vehicle movement planning.
Why it matters
By providing open models that run on controlled infrastructure, NVIDIA enables organizations to deploy specialized AI using internal data rather than relying on external, cloud-hosted black-box APIs. This shift prioritizes data sovereignty and vertical-specific applications across domains from finance to biology.
The release includes the Cosmos 3.5 model for robotics and BioNeMo RFD4-Proteina, the latter being the fourth iteration of NVIDIA’s protein generation series. These models are designed to run locally on controlled hardware rather than relying on centralized cloud compute.
The players
NVIDIA
A designer of graphics processing units, AI accelerators, and open-source model architectures for specialized enterprise and scientific computing.
Coinbase
A digital asset exchange platform that has adopted the Kumo model family to process structured business data for risk and forecasting.
The details
The new models, including Kumo for structured business data, are designed for local adaptation using proprietary internal datasets. Alpamayo 2 Edge—a specialized model for interpretation of real-time driving scenes—functions by processing visual inputs through onboard computing hardware to plan vehicle movement. BioNeMo RFD4-Proteina uses similar local-execution architecture to generate protein sequences without requiring data to leave the user's secure infrastructure.
Timeline
April 2026: NVIDIA introduced the Ising quantum computing model family.
October 21, 2026: NVIDIA released the Kumo, Nemotron Retriever, Cosmos 3.5, and BioNeMo RFD4-Proteina models.
Late October 2026: Alpamayo 2 Edge is scheduled for release.
The Tech Race
This release expands the existing BioNeMo research ecosystem, positioning NVIDIA's open models against proprietary alternatives in the scientific and enterprise sectors. It follows the trend of modular, vertical-specific AI architectures, similar to the company's prior work with the Ising quantum family.
Organizations can immediately utilize Kumo and BioNeMo RFD4-Proteina on their internal infrastructure to begin forecasting or sequence generation tasks. Enterprise users and developers should prepare to integrate these models into their local data pipelines once downloads become available.
The takeaway
NVIDIA is moving toward an increasingly granular model ecosystem that allows companies to maintain control over sensitive data while utilizing state-of-the-art AI. Watch for performance metrics to be published following the release of the Alpamayo 2 Edge model in late October.
What happens next
Users can look forward to the release of the Alpamayo 2 Edge model, which is scheduled for download in late October 2026.
Further reading
Explore the ongoing shifts in how organizations deploy proprietary workloads at Artificial Intelligence.
Source note: This article includes information reported by TokenPost.
Live Poll
Should businesses trust third-party artificial intelligence models for making core predictions and operational decisions?
