Perplexity drops Decider V1.3 on OpenRouter

Treating general-purpose LLMs as classifiers is increasingly an anti-pattern when high-throughput routing demands determinism, speed, and cost efficiency. Purpose-built decision engines like Decider bypass text parsing entirely by returning calibrated probability distributions straight to application logic.

• –Pure probabilistic output eliminates JSON parsing failures and hallucinations across boolean decisions, 255-way categorical choices, and multi-tier rubric scores.
• –Span extraction via the new locate method pulls exact text and list values with character offsets and confidence ratings.
• –Multimodal support across text and image states enables fast visual classification within a 262K context window before delegating to heavier models.
• –Perplexity's decision models now process 235M requests monthly on OpenRouter, capturing over 15% of the platform's decision traffic.