“Groq's solution provides inference at approximately 1/10th the cost per token compared to a modern GPU.”