HalluWorld - Grid Memory: leaderboard
Metric: Hallucination rate (%) on the memory probes (tracking states across earlier observations, 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. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 3.8 |
| 2 | Claude Opus 4.6 (Medium) | 3.8 |
| 3 | O4 Mini | 4.5 |
| 4 | O3 Mini | 4.9 |
| 5 | Claude Sonnet 4.6 | 5 |
| 6 | O3 | 5.5 |
| 7 | GPT-5.5 (Medium) | 5.9 |
| 8 | Kimi K2.6 (Non-reasoning) | 6.2 |
| 9 | Claude Sonnet 4.6 (Medium) | 6.2 |
| 10 | DeepSeek V3 (0324) | 14.3 |
| 11 | GLM-5 (Non-reasoning) | 15.8 |
| 12 | GPT-4o | 17.2 |
| 13 | GPT-5.4 Mini (Non-reasoning) | 26.6 |
| 14 | Qwen 3 30B A3B (Non-reasoning) | 27.7 |
| 15 | GPT-4o Mini | 39 |
Interactive version: theaggregate.ai/benchmark?slug=halluworld-grid-memory · How It Works · Data refreshed daily, snapshot 2026-10-07.