Perplexity releases open-weight Decider v1.3 and Decider Flash v1.3

Perplexity AI released version 1.3 of its open-weight Decider decision models on November 12, 2026, adding a span-extraction feature called locate and publishing the model weights on Hugging Face, OpenRouter, and its own Decisions API, according to Perplexity Decider.

What changed in version 1.3

• Decider v1.3 is a fine-tune of Qwen 3.8-27B, and Decider Flash v1.3 is a distilled fine-tune of Qwen 3.5-4B, succeeding Decider v1.2 and Decider Flash v1.2
• Both models are released under the Apache 2.0 license as downloadable weights on Hugging Face
• The models are available through the Perplexity Decisions API, on Hugging Face, and via OpenRouter
• A new locate feature extracts text spans for named entity recognition, PII detection, and extractive question answering
• Version 1.3 adds bidirectional attention in its full-attention layers and supports binary and multi-class classification with probability outputs
• Decider v1.3 has a context window of 262,144 tokens

What it means for people who build with AI

Perplexity reports that Decider v1.3 reaches 91.6% mean accuracy on query classification, 6.54 points above the prior production model, 83.99 nDCG@10 across 72 retrieval and reranking tasks, 88.16% mean accuracy across 8 languages, and 93.98% judge agreement on real-world API requests.

Decision models that output structured classifications or probabilities instead of free text are becoming a distinct, cheaper model category that AI coding and agent tools can plug in for routing, filtering, and extraction tasks instead of calling a full LLM.

Sources

• Perplexity Decider (Perplexity AI, November 12, 2026, official)
• We're rolling out pplx-decider-v1.3 and pplx-decider-flash-v1.3 (Perplexity API Platform Forum, November 12, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.