Mouse-1 (mouse.dev)

Today, a new decision model called Mouse-1 was announced, aimed at enhancing AI agent harnessing capabilities. It integrates through a single API, providing decisions in approximately 100 milliseconds and boosting the independence and reliability of AI agents. Notably, Mouse-1 exhibited a staggering 163× speed advantage over GPT-6 Sol, producing decisions before Sol could even generate its first token. In rigorous testing, Mouse-1 achieved an average decision accuracy of 89.1%, outperforming several competitors, thus signaling a significant leap in model efficiency and effectiveness. The introduction of Mouse-1 carries substantial implications for the AI/ML community, particularly in decision-making applications. Its enhanced resilience against prompt injection attacks, achieving a 0% alteration in decisions under planted instructions, showcases advancements in security alongside performance. The model's ability to process image inputs alongside text further broadens its utility in diverse scenarios. Priced competitively at $0.10 per million input tokens with free output, Mouse-1 not only offers improved cost-effectiveness but promises to streamline operations for organizations relying on automated decision-making, refining the workflow of AI-powered agents across various industries.