TacitVLA is a new vision-language-action model that improves performance and reduces latency by generating actions independently of reasoning text. It outperforms previous results on various benchmarks and has the potential to enhance AI agent capabilities.