AdaExploreBench (Stationary) - Calibrated Accuracy: leaderboard

AdaExploreBench's calibrated accuracy with the rule drift removed, a diagnostic setting: the agent still has to discover the hidden decision rule with its own probes, but the rule never changes, so the gap to the main setting measures the cost of undetected drift. Gemini-3.8-Flash and GPT-6-Sol have no runs here.

Metric: CA (%) (0-100). Source: hkust-knowcomp.github.io. Saturation forecast: Around February 2027. 5 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1DeepSeek V4.1 FlashHigh67.6#69
2Qwen 3.8 27BMedium50.57#220
3GPT-6 LunaHigh41.14#106
4Gemma 4 12B (IT)High22.78#731
5Qwen 3.5 9B12.41#829

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=adaexplorebench-stationary-calibrated-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-20.