Perplexity ships pplx-decider-v1.1-27b, an open-weights decision model at half v1's API price

Released Oct 6, the Qwen3.8-27B-based model returns probabilities for yes/no, choice, and rubric-score questions instead of text, lifting its Hugging Face Decision Index score from 56.4 to 61.56 and costing $0.02 per 1M input tokens via POST /v1/decisions. Self-hosting the weights needs roughly 49 GiB of GPU memory plus Perplexity's bundled inference code, since default causal inference doesn't reproduce the evaluated setup.