pplx-decider v1.1 leads EdgeEvals' clinical preview, with one severe error

Source · Model card: pplx-decider v1.1 27B

The 11 October EdgeEvals System One preview credits pplx-decider v1.1 on 643 of 669 decisions. It has the highest credited count and shares rank 1 with GPT-6 Luna, which has 637. Mercury Decide has 630 credits and Jev 1.13 has 628; both are ranked second. The board records one severe error for pplx-decider and Jev, none for GPT-6 Luna, and two for Mercury.

The published JSON records a 362 ms median and an API cost of $0.02543 per 1,000 decisions, or about $0.017 for all 669. That matches the post’s claim that the full run cost less than two cents.

The board is a preview built from cases authored by its operator. Its methodology says clinicians have not reviewed them, so this is not clinical validation. The post says eight models joined since Thursday; the current JSON has 35 rows, while the 10 October Microsoft snapshot we previously recorded had 38. The roster change is unexplained, so we cannot confirm a net increase of eight.

The result adds a dated benchmark note to pplx-decider v1.1. Its catalog score stays unchanged.

Sources: board · data · methodology · @MaziyarPanahi.