HalluWorld - Grid Perceptual: leaderboard

Perceptual hallucination in HalluWorld-Grid (2026), MiniGrid worlds whose ground truth comes from the environment state: the rate of hallucinated answers to questions about values present in the current observation. Percent; lower is better.

Metric: Hallucination rate (%) on the perceptual probes (read-out of values present in the current observation, 6 levels) 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
1O30.3#237
2O3 Mini0.4#546
3GPT-5.5Medium0.4#54
4O4 Mini0.8#368
5Claude Sonnet 4.6Medium1.1#173
6Claude Opus 4.6Medium1.7#113
7Kimi K2.6Non-reasoning5.4#192
8GPT-4o10#751
9GLM-5Non-reasoning19.8#269
10GPT-5.4 MiniNon-reasoning19.8#406
11Qwen 3 30B A3BNon-reasoning22.4#1156
12DeepSeek V3 (0324)23.5#715
13GPT-4o Mini34.9#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-perceptual · How It Works · Data refreshed daily, snapshot 2026-11-21.