pplx-decider-v1-27b

Not yet independently verified. The date is the Hugging Face repository’s creation time (1 October 2026, 21:24 UTC). The accuracy comparison is Perplexity’s own evaluation, so no score is stored; the only other mention found is a newsletter roundup. We will update this when it can be confirmed, and remove this note.

pplx-decider-v1-27b is an open-weight decision model published by Perplexity on Hugging Face on 1 October 2026 under the Apache 2.0 licence. It is a fine-tune of the Qwen3.8-27B base model with about 27 billion parameters and a context of 262,144 tokens, and its model card tags it for text classification. Decision models of this kind answer constrained questions, picking one of the options supplied with the prompt or giving a judgement, instead of generating open-ended text, which makes them suited to routing requests, checking outputs and gating actions inside agent systems. Perplexity’s card compares it with other decision models across eleven benchmarks and reports a higher overall accuracy; those are its own results, measured through its API, so no score is stored here. It joins several decision models released at the start of October, including the Clef models and Strands Decider 2B already on the board. The weights are free to download; no hosted price is stated.