Perplexity Decider
by Perplexity · Model · Language & code models
mixed
Open-weight extraction model that some users like, but results against v1.2 are split
pplx-decider is a 27B decision model from Perplexity, fine-tuned from Qwen3.8-27B
AccessOpen weightsNeeds~17 GB (the Ollama build)Downloadabout 17 GB through OllamaLicenseApache 2.0
Why we say so
Perplexity claims it beats production classifiers by 6.54 points on 63 tasks, but that is the maker's own number. HN commenters split: one found it beat Jev 1.14 and v1.2 on an internal extraction eval, another found it a step back from v1.2 for their use case. One commenter also found ScaleDown Extract cheaper and more accurate than it.
Evidence
- Nov 13Open-weight Decider v1.3 adds 'locate' span offsets for NER, PII, extractive QA; 6.54 points above production classifiers on 63 tasks; free on OpenRouter one week
- Nov 13Beat Jev 1.14 and Decider v1.2 on an internal entity extraction/matching eval
- Nov 13Priced competitively and did well on a commenter's local benchmark; seen as a replacement after Inception discontinues the Mercury API
Cost to run
Alternatives
More Perplexity Decider
Built on Perplexity Decider