Review Error Injection (OpenAIReview): leaderboard

Metric: Error detection recall (%; share of injected errors that the review's comments quote and correctly explain, fuzzy quote match then a Gemini-3 Flash Preview explanation judge) on the 24-paper frontier subset of a perturbation benchmark built from 74 clean arXiv papers in 8 subject classes, with surface (math-token) and claim, reasoning and experimental errors injected; backends via OpenRouter without reasoning mode; the OpenAIReview agentic review system (passage-level review with a validate-and-dedupe pipeline) backed by each model; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.

Top models

#ModelScore
1OpenAIReview + GPT-5.5 (Non-reasoning)71.6
2OpenAIReview + Claude-Opus-4.768.1
3OpenAIReview + DeepSeek-V4-Flash (Non-reasoning)55.8
4OpenAIReview + Grok-4.1-Fast (Non-reasoning)52.6
5OpenAIReview + Qwen3.6-35B-A3B (Non-reasoning)45.4
6OpenAIReview + Gemini-3.1-Flash-Lite (Non-reasoning)33

Interactive version: theaggregate.ai/benchmark?slug=review-error-injection-openaireview · How It Works · Data refreshed daily, snapshot 2026-09-29.