Benchmark results

Per-task results

Every prediction behind the leaderboard, item by item.

All predictions on HuggingFace Browse items

Item explorer

What each model answered

Pick a task. For each item, the page groups every configuration by its answer. Green is the correct answer.

Grading: the answer inside the last <answer>…</answer> tag is compared by exact match after separator-insensitive normalisation (see scoring.py). “No answer” means the model returned no parsable answer or the request failed.