Daily Tokens
November 13, 2026

AI · November 13, 2026 · Moderate impact

Perplexity releases Decider v1.3 (26B) and Decider Flash v1.3 (5B) as open-weight classifiers under Apache 2.0

Perplexity has released two decision models that return probability distributions over answer options for a given input state and question, rather than generating text token by token. The 26B pplx-decider-v1.3 is fine-tuned from Qwen3.8-27B and needs about 49 GiB for BF16 weights. The 5B pplx-decider-flash-v1.3 is built on Qwen3.5-4B and distilled from the larger model using a KL objective on the teacher's distributions, needing about 8.5 GiB in BF16. Both support text and images and are licensed under Apache 2.0.

Open-weight classifiers built on LLM backbones give developers a deployable way to score and route inputs without paying for generative output on every call.

Go deeper

More from AI on Friday, November 13