Question Answering
- AI2 ARC-Challenge (test) · subset · 2 measured runs
- OpenBookQA main (test) · subset · 2 measured runs
- ARC-Easy · subset · 11 measured runs
- ARC-Challenge · subset · 11 measured runs
- OpenBookQA · subset · 11 measured runs
- ARC-Easy (full test split, raw accuracy) · full · 11 measured runs
- ARC-Easy / Qwen3.5 same-base cohort · full · 15 measured runs
- ARC-Challenge / Qwen3.5 same-base cohort · full · 15 measured runs
- ARC-Challenge (full test, raw likelihood accuracy) · full · 11 measured runs