PaperAudit-Bench (Fast Mode) - o4-mini Errors: leaderboard

Metric: Macro-F1 (%; the per-paper harmonic mean of error coverage and finding precision, averaged over the 46 NeurIPS 2025 oral papers of the benchmark's NeurIPS branch as corrupted by o4-mini with 10-20 injected errors of eight types; Fast mode, one global review pass over the whole paper; findings matched to the injected errors by a GPT-5.1 judge; the three text-only detectors read the papers with figures removed). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.

Top models

#ModelScore
1Grok 445.2
2Gemini 2.5 Pro44.6
3GPT-542.1
4Qwen 3 235B A22B 2507 Instruct40.5
5O4 Mini39.2
6GLM-4.6 (Non-reasoning)38.7
7Claude Sonnet 4.536.1
8DeepSeek V3.1 (Non-reasoning)23.2
9Kimi K2 090514.4

Interactive version: theaggregate.ai/benchmark?slug=paperaudit-bench-fast-mode-o4-mini-errors · How It Works · Data refreshed daily, snapshot 2026-09-29.