ReactBench (Hallucination) - Alteration Tracing (Standard): leaderboard

Accuracy in percent on ReactBench (2026) questions about before-and-after image pairs, asking what changed, with standard prompting; an incorrect answer counts as a hallucination. Higher is better.

Metric: Accuracy (%) on Alteration Tracing (10,956 questions on 913 before-and-after image pairs) 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 February 2027. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3.5 27B79.9#380
2Qwen 3 VL 32B Instruct73.5#607
3Qwen 3.5 9B67.4#838
4InternVL3-8B64.1#1795
5Qwen 2.5 VL 72B Instruct62.1#849
6Qwen 3 VL 4B58#1163
7Qwen 2.5 VL 7B Instruct54#1919
8InternVL2.5-26B51.3#2174
9InternVL2.5-8B45.4#2393
10ReactBench LLaVA-OV-7B (checkpoint unspecified)30.6
11ReactBench LLaVA-v1.6-7B (checkpoint unspecified)27.3

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-alteration-tracing-standard · How It Works · Data refreshed daily, snapshot 2026-11-22.