Nvidia Releases Nemotron 3.5 Open Models, Claims Enterprise Agent Performance Beats OpenAI and Anthropic

Nvidia launched a new series of open-weight models on October 19, 2026, including the Nemotron 3.5 Super VL for enterprise-grade agents and lightweight Bolt models for personal computers and workstations. The Super VL features 120 billion parameters, and Nvidia claims it delivers accuracy comparable to its larger Ultra model in instruction-following and tool-use tasks while being more efficient.

Addressing market competition, Nvidia stated that Super VL completes tasks faster than proprietary models like Anthropic's Opus 5 and OpenAI's GPT 6.1 Sol when post-trained on specific agentic tasks such as document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience. These performance comparisons are sourced from Nvidia's official blog and represent vendor self-tests, lacking independent third-party reproduction.

Beyond cloud and enterprise data center scenarios, Nvidia introduced the Bolt 74B-A9B and Bolt E4B small models, specifically designed to run locally on new AI PCs equipped with RTX Spark chips or DGX Spark workstations. This move aims to attract consumers and enterprise knowledge workers who prioritize privacy and wish to delegate everyday projects to local agents without uploading workloads to the cloud.

Kari Briski, Nvidia's Vice President of Open Models and Software, stated that organizations can now deploy and optimize these open models on their own infrastructure without sacrificing capability. Despite competition from non-Nvidia chips like AWS Trainium and Cerebras, Nvidia is strengthening its position in enterprise AI infrastructure by providing an optimized open-source software stack.