With today's launch of NVIDIA Nemotron 3.5 Super VL, we're collaborating with NVIDIA to help enterprises customize Nemotron open models for the work that matters most to them. We've had early access to the model and have been post-training it ahead of the launch with enterprises like Skydance and EY.
Starting today, this service is available to any enterprise through Scale Post-Training, a new offering within Scale GenAI Portfolio that helps enterprises customize leading open-weight models, including Nemotron 3.5 Super VL and the full Nemotron open model family. Post-training teaches an existing model a particular job using examples of that job done well, turning a general-purpose model into a specialist. For Nemotron, we combine our evaluation methodology, data and training expertise with NVIDIA post-training tools, so enterprises can balance cost, performance and control in building agentic systems and workflows that run natively in their environments.
This collaboration advances our mission to develop reliable AI systems for the world's most important decisions. For more than a decade, our data and evaluations have helped train the world's leading AI models. Now, we're bringing that same work to enterprises customizing open-weight models.
Turning Open Models Into Enterprise Specialists
The question for enterprises is no longer which model is best in general. It's which model is best for a specific job, at a sustainable cost and speed for the business. At production volume, inference costs add up quickly. Deployments will increasingly combine multiple models of different sizes, modalities, providers, strengths and costs, and open-weight models are a growing part of that mix – especially when paired with expert post-training in service of a specific use case or workflow. They can run inside a company's own infrastructure and be tuned to specific tasks, which gives enterprises more control over quality, cost and where their data is processed.
“Open models give enterprises a head start on building with their own data and intelligence,” said Joey Conway, senior director of Nemotron Models and Software at NVIDIA. “With NVIDIA Nemotron open weights, datasets and post-training tools, Scale AI is creating specialized models that boost accuracy with custom business intelligence, powering efficient systems of models for advanced agentic systems and workflows.”
"NVIDIA is giving enterprises open models they can actually make their own, and Nemotron is a big step forward,” said Francis deSouza, CEO of Scale AI. “The future of enterprise AI is multi-model, and the advantage will go to organizations that can prove which model is best for each job. With NVIDIA, Skydance and EY, we’re showing that with Scale Post-Training, an open-weight model can match or outperform the models enterprises run today at a fraction of the cost."
Every general-purpose model – open-weight or closed – starts as a generalist that doesn't know a company's policies, terminology or workflows. Closing these gaps takes more than access to the weights. Someone has to define what "correct" means for each workflow through a representative evaluation dataset, build the data that teaches the model and prove the system is ready for production. This is the work Scale does. We measure the model a customer runs today on tasks from their own workflow, build training and evaluation data for that workflow, find the right post-training recipe, post-train Nemotron with NVIDIA's open tools and test each version inside the same agent workflow it will run in production. Along the way, our engineers contribute improvements back to those tools for the open-source community. Read more about our NVIDIA partnership here.
Enterprises Building With Nemotron
With Skydance, we’re using Nemotron 3.5 Super VL to automate a core finance workflow. The agent reviews vendor credit memos, statements and invoices, many of them scanned, and matches them against the company’s financial records to find credits it’s owed. Tested on real tickets from the production workflow, Nemotron post-trained by Scale matched the accuracy of the frontier model Skydance runs in production today, at about 40x lower inference cost and 5x faster response times.
With EY, we’re building a due diligence agent that answers deal questions by finding and reconciling information across the hundreds of files in an M&A virtual data room. In evaluations across hundreds of questions from held-out data rooms, Nemotron 3.5 Super VL post-trained by Scale matched the accuracy of the frontier model EY uses today, at about one-eighteenth of the inference cost and with answers returned more than twice as fast.
We're also working with a global insurance company on a copilot that helps customer service representatives during live French-language roadside assistance calls, where getting the answer right the first time builds customer trust and keeps costs in check. The copilot pulls out the key facts of each case, recommends the benefits the customer's policy covers and flags when a representative might promise something that policy doesn't include. In controlled evaluations, Nemotron 3.5 Super VL with Scale Post-Training was up to 31 percentage points more accurate than the model the company uses in production today, at about a quarter of the inference cost.
Get Started With Scale Post-Training
Scale Post-Training is available now for enterprises that want to customize open-weight models for their own workflows, including Nemotron 3.5 Super VL.