HalluWorld - Grid Compound: leaderboard

Compound hallucination in HalluWorld-Grid (2026), MiniGrid worlds whose ground truth comes from the environment state: the rate of hallucinated answers to compound questions that combine several probe categories. Percent; lower is better.

Metric: Hallucination rate (%) on the cross-category compound probes of HalluWorld-Grid (MiniGrid worlds whose ground truth comes from the environment state), micro-averaged over levels and serializers, temperature 0 where supported, thinking models at medium effort with up to 16k thinking tokens, every model at most 256 answer tokens; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.5Medium2.7#54
2Claude Opus 4.64.6#113
3O35.7#237
4O4 Mini5.8#368
5Claude Sonnet 4.6Medium7.5#173
6O3 Mini7.6#546
7Kimi K2.6Non-reasoning8.8#192
8GPT-4o11.3#751
9DeepSeek V3 (0324)12.7#715
10GLM-5Non-reasoning13.4#269
11Qwen 3 30B A3BNon-reasoning23#1156
12GPT-5.4 MiniNon-reasoning24#406
13GPT-4o Mini28.7#1667

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=halluworld-grid-compound · How It Works · Data refreshed daily, snapshot 2026-11-21.