PaperAudit-Bench (Fast Mode) - Claude Sonnet 4.5 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 claude-sonnet-4-5-20250929 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 September 2027. 10 models tracked.

Top models

#ModelScore
1Claude Sonnet 4.547.3
2GPT-538.7
3GLM-4.6 (Non-reasoning)38.1
4Qwen 3 235B A22B 2507 Instruct36.2
5Gemini 2.5 Pro35.4
6Grok 435.1
7O4 Mini30.7
8Kimi K2 090527.4
9DeepSeek V3.1 (Non-reasoning)26.2

Interactive version: theaggregate.ai/benchmark?slug=paperaudit-bench-fast-mode-claude-sonnet-4-5-errors · How It Works · Data refreshed daily, snapshot 2026-09-29.