Perplexity Decider v1.3: Training, eval, and use cases

Perplexity's Decider v1.3 reached 90% accuracy at judging whether a query can be answered by a retrieval-based system, the company reported in a blog post documenting the model's training, evaluation and use cases. Perplexity positions Decider v1.3 as a decision layer rather than a general answer model, and says routing simple queries away from costly LLMs can cut that reliance by 4x, yielding savings for production environments serving millions of user queries daily. Perplexity advises teams to treat Decider v1.3 as a candidate replacement for brittle routing heuristics but to validate it on their own traffic, since bad decision-policy calls cascade into tool choice, cost and answer quality.

Hacker News https://www.perplexity.ai/hub/blog/perplexity-decider Perplexity Decider v1.3: Training, eval, and use cases Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated. Perplexity's Decider v1.3 achieved 90% accuracy in determining whether a query can be answered by a retrieval-based system, potentially reducing reliance on costly LLMs for simple queries by 4x, enabling significant cost savings for production environments serving millions of user queries daily. Perplexity has moved Decider to v1.3 and is documenting its training, evaluation, and use cases as a decision layer rather than a general answer model. For production agent stacks, the practical takeaway is to treat it as a candidate replacement for brittle routing heuristics, but validate it on your own traffic because bad decision-policy calls will cascade into tool choice, cost, and answer quality.