SpatialText (Code-Generated) - 2D Non-Omniscient: leaderboard
Metric: Accuracy (%) on 2D scenes whose description omits relations, so some answers are undecidable (400 questions) on SpatialText's programmatically generated structured scenes (free-response, yes/no, choice and query-set questions), text only, greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 8B | 80 | #667 |
| 2 | Gemma 3 12B (IT) | 59.8 | #655 |
| 3 | DeepSeek R1 Distill Llama 8B | 55 | #1282 |
| 4 | Gemma 2 9B (IT) | 49.8 | #774 |
| 5 | Qwen 2.5 7B Instruct | 39 | #846 |
| 6 | Mistral Instruct v0.1 7B | 30.5 | #1429 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=spatialtext-code-generated-2d-non-omniscient · How It Works · Data refreshed daily, snapshot 2026-10-11.