Eon's Era Achieves Frontier-Level Accuracy with NVIDIA Nemotron 3.5 Super VL

WHAT HAPPENED

Eon's Era platform achieved 88.9% accuracy using NVIDIA Nemotron 3.5 Super VL at 27 times lower cost per correct answer than frontier models. The company then used Era-generated data to post-train Nemotron for $609, improving the model's ability to recognize missing information while reducing inference costs further.

WHO IS INVOLVED

Eon partnered with NVIDIA to test Nemotron 3.5 Super VL using the Era platform. Ofir Ehrlich, representing the Eon team, announced the collaboration and results on LinkedIn.

WHY IT MATTERS

The results demonstrate a workflow that closes the loop between benchmarking and post-training, allowing enterprises to identify model failures, generate training data to fix them, and test improvements at low cost. This could change how companies customize open-source models for their specific environments.

Read more

• linkedin.com ↗