Python experiment harness for comparing modular prompt strategies on factual questions, with a SimpleQA case study.
-
Updated
Sep 22, 2026 - Python
#
simpleqa
Here are 4 public repositories matching this topic...
Proper scoring rules, reduces LLM overconfidence in multiple-choice QA.
calibration hallucination confidence-estimation abstention selective-prediction llm mmlu proper-scoring-rules mmlu-pro simpleqa
-
Updated
Jun 13, 2026 - Python
SimpleQA benchmark of Keiro deep search: 96.5% on 200 unseen questions, 94% in ~900 tokens of evidence. Runner and every result included.
-
Updated
Oct 30, 2026 - Python
Open SimpleQA benchmark of Keiro Lite search: 92.0% with an open, no-model recipe. Crawler, judge and every result included.
-
Updated
Oct 28, 2026 - Python
Add this topic to your repo
To associate your repository with the simpleqa topic, visit your repo's landing page and select "manage topics."