AI

Scale AI and NVIDIA Launch Enterprise Nemotron Post-Training Service

Thursday, October 15, 2026Read Original

Details

  • Scale AI launched Scale Post-Training on October 15, 2026, enabling enterprises to customize NVIDIA’s Nemotron 3.5 Super VL and broader Nemotron open-model family for specific workflows.
  • The collaboration combines Scale’s evaluation, data, and training expertise with NVIDIA’s open post-training tools, helping companies optimize model quality, inference cost, deployment speed, and data control.
  • Nemotron 3.5 Super VL is an open 120-billion-parameter multimodal model for reasoning, tool use, and text, image, document, and video understanding; NVIDIA says it activates 12 billion parameters per token and supports deployment from data centers to edge hardware.[1][5]
  • Scale reported that post-trained Nemotron matched production frontier-model accuracy for Skydance at roughly 40-times lower inference cost and five-times faster responses, while an EY due-diligence agent matched accuracy at about one-eighteenth the cost and more than twice the speed.
  • A global insurer’s French-language roadside-assistance copilot was reportedly up to 31 percentage points more accurate than its existing production model at approximately one-quarter of the inference cost.

Impact

The launch strengthens the enterprise case for open-weight, multimodal models that can be specialized and deployed within company-controlled environments. Over the next 12–24 months, demand is likely to shift from selecting a single general-purpose model toward evaluating multi-model systems by workflow economics, reliability, and governance, increasing investment in post-training data, testing, and agent infrastructure.

Rift Dispatch