ABAB News Logo
Back to Flash News

Perplexity open-sources Decider v1.3 decision models and adds locate functionality

Perplexity co-founder and CEO Aravind Srinivas posted on X on November 12 that the company has open-sourced its latest decision models, Decider v1.3 and Decider Flash v1.3, and is offering them for free on OpenRouter for one week.

v1.3 adds a “locate” feature, allowing users to describe the content they want to find, after which the model returns matching text and character offsets. Srinivas said the feature is suitable for batch decision-making and evidence/entity extraction; on 17 out-of-domain NER, PII, and extractive QA tasks, the Decider models significantly outperformed the best general-purpose extraction model. The company said that on 63 proprietary production classification tasks, v1.3’s accuracy was 6.54 percentage points higher than that of the existing classifier. The weights have been made available on Hugging Face.

Original source:

  1. Title: Aravind Srinivas X post (open-sourcing Decider v1.3)
    Publisher: Aravind Srinivas (@AravSrinivas)
    Publication date: 2026-11-12
    URL: x.com/AravSrinivas/status/2121020114321703210
    Source type: original post by a company executive
    Verification status: the post text was actually read in this case, verification date 2026-11-12
    Supporting content: supports the open-sourcing of Decider v1.3 and Flash v1.3, one week free on OpenRouter, the locate feature, and the performance statements

ABAB AI Insight

Decision models output probabilities rather than generating text, making them suitable for structured tasks such as classification, scoring, and extraction. The locate feature directly incorporates text-span extraction into the same API, reducing post-processing steps.

Open-sourced weights and a limited-time free offering lower the barrier to trial, while the company retains a paid API path. The performance comparison is based on the company’s proprietary tasks and out-of-domain benchmarks; the scope of independent replication remains to be seen.

ABAB News · Cognitive Laws

  1. Decision models can combine text-span extraction and probability outputs in the same request.

  2. Open-sourced weights together with a limited-time free API accelerate external validation and adoption of models for specific tasks.