Can Decision Models Replace LLMs in Real-World Contact Center Classification and Judging Tasks? A study posted as arXiv:2611.22949v1 found that many decision models match frontier generative LLMs on contact center classification tasks, with API-based closed decision models achieving the lowest per-request cost among the API models evaluated. The paper compared nine decision models against thirteen generative LLMs in zero-shot settings across eight contact center use cases — six classification tasks and two judging tasks — and found decision models' main weakness is abstention, rarely choosing the no-answer option, which accounts for most of their gap to the strongest LLMs, with Gemini-Decision-1 the notable exception. The authors report that decision models' confident predictions are more accurate, enabling selective routing in production, but their judging quality drops sharply for smaller models and they are more prone to position bias, leading the authors to recommend selecting decision models on abstention behavior, confidence and deployment cost rather than overall accuracy alone. arXiv:2611.22949v1 Announce Type: new Abstract: Contact center systems often need to make bounded decisions, such as identifying a caller's intent, retrieving the knowledge-base article that answers a query, assessing agent compliance, or judging the output of AI models used in production. Decision models return a typed answer and its probability directly, making them a promising cost-effective alternative to generative Large Language Models LLMs in such bounded conversational tasks. However, their effectiveness in real-world contact center applications remains unclear. In this paper, we compare nine decision models against thirteen generative LLMs in zero-shot settings across eight contact center use cases: six classification tasks and two judging tasks. Our experiments demonstrate that many decision models can match the performance of frontier LLMs on contact center classification, and API-based closed decision models achieve this at the lowest per-request cost among the API models we evaluate. We further observe that their confident predictions are more accurate, enabling selective routing in real-world production scenarios. Across classification tasks, the main weakness of most decision models is abstention: they rarely choose the no-answer option, which accounts for most of their gap to the strongest LLMs although Gemini-Decision-1 is the notable exception in this regard . We also find that the judging quality of decision models drops sharply for smaller models, and they are also more prone to position bias. In short, our findings provide practical guidance for selecting decision models based on abstention behavior, confidence, and deployment cost, rather than overall accuracy alone.