ReactBench (Hallucination) - Relational Erasure (Standard): leaderboard

Accuracy in percent on ReactBench (2026) questions about images where an object or relation was edited out, with standard prompting; an incorrect answer counts as a hallucination. Higher is better.

Metric: Accuracy (%) on Relational Erasure (19,140 questions on 1,595 images where an object or relation was edited out) with standard prompting, hallucination-inducing questions (multiple choice, short answer, yes/no) about images edited by Qwen-Image-Edit-2511; an incorrect answer counts as a hallucination; temperature 0.2, at most 1,024 tokens, LLM answer matching; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 32B Instruct64.5#607
2Qwen 2.5 VL 72B Instruct55.6#849
3InternVL3-8B51.6#1795
4Qwen 3 VL 4B50.1#1163
5InternVL2.5-26B48.3#2174
6Qwen 3.5 27B48#380
7Qwen 2.5 VL 7B Instruct47#1919
8Qwen 3.5 9B43.2#838
9ReactBench LLaVA-OV-7B (checkpoint unspecified)42.1
10InternVL2.5-8B38.6#2393
11ReactBench LLaVA-v1.6-7B (checkpoint unspecified)29.3

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=reactbench-hallucination-relational-erasure-standard · How It Works · Data refreshed daily, snapshot 2026-11-22.