PT-BR Typed-Decision Benchmark with 12 External Models

Models select typed answers from options; the main metric is balanced accuracy with shuffled choices and confidence intervals. Evaluation runs locally, not on hardware benchmarks or general chat performance. Participants must submit full reports with pinned checkpoints, runtime settings, and validation outputs. Saracura models are excluded from the main ranking due to task-specific fine-tuning and shown separately.

Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.