Yesterday
-
Robustness tab: 19 models (adds GPT-6.1 Terra, Qwen3.5-122B, DeepSeek-R1, phi-4, GLM-4.6)
-
Add Robustness tab: persuasion, alarm and wrapper tracks for 14 models (2026-11-14)
3 days ago
-
Add GPT-6.1 flagship (rank 1, bal 99.5); closed-weight reference
-
Restore Darwin-28B to board (holdout = official split); 31B internal
-
Scoring note: public dev set is reference; official score on private holdout only
-
Hold Darwin from public board (dev-vs-holdout gap not reproducible)
4 days ago
-
Add Darwin-28B-Opus (rank 7, bal 86.6); keep Darwin-31B internal
-
Scatter: numbered markers + full name legend (no overlapping text, every model named)
-
Declutter scatter labels (3 anchors + hover tooltips)
-
Add Analysis tab: scatter (catch vs allow), obfuscation-gap bars, benign-pass bars; +phi-4
-
Update board: +DeepSeek-R1; Darwin held off public (chairman 11-12)
-
Upload index.html with huggingface_hub
