SpatiaLQA - Content F1: leaderboard
Metric: Step-content F1 (%, 0-100): harmonic mean of the pooled recall and precision of matched step contents; SpatiaLQA: 9,605 questions on 2,401 photographs of 241 real indoor scenes (13 categories), each asking for the ordered steps of a multi-step task and every step's preconditions in a fixed JSON template; GPT-4o matches predicted to annotated steps per image (a 0/1 matrix filtered by the Hungarian algorithm), then recall and precision are pooled over all samples; open models with their Hugging Face recommended settings, proprietary models at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 41 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 76 | #91 |
| 2 | Gemini 2.5 Pro | 74.3 | #145 |
| 3 | GPT-5 Mini | 73.9 | #176 |
| 4 | GPT-4.1 | 73.5 | #240 |
| 5 | Gemini 2.5 Flash | 72.9 | #237 |
| 6 | Qwen 2.5 VL 72B Instruct | 69.6 | #364 |
| 7 | Claude Sonnet 4 | 68.9 | #194 |
| 8 | GPT-4o | 67.4 | #333 |
| 9 | GPT-4.1 Mini | 66.5 | #346 |
| 10 | GPT-4o Mini | 63.5 | #588 |
| 11 | Claude 3.7 Sonnet | 59.3 | #241 |
| 12 | GLM-4.1V-9B (Thinking) | 57.5 | #457 (GLM-4.1V-9B) |
| 13 | Pixtral-12B | 57 | #795 |
| 14 | Qwen 2.5 VL 32B Instruct | 54 | #443 |
| 15 | Qwen 3 VL 8B Instruct | 52.7 | #401 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=spatialqa-content-f1 · How It Works · Data refreshed daily, snapshot 2026-10-11.