NC AI announced VAETKI Vision 3B, a roughly three-billion-parameter vision-language model. The company reports an average score of 63.8 across three robotics and spatial-understanding benchmarks, ahead of six compared open models, and says it processes eight frames from four seconds of video together. A welding-specific variant improved one classification score from 41.1 to 80.9. These are limited company-reported evaluations, not proof of long-term production stability, latency, energy efficiency or safe real-world task performance.