ST-BiBench - Foundational Spatial Grounding: leaderboard
Metric: Success score (%) for assigning the reachable arm and target in bimanual scenes, the mean of the sparse, dense and cluttered (with distractors) scenario settings, at least 100 domain-randomized episodes each; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 29 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash | 95.38 | #331 |
| 2 | Gemini 2.5 Flash | 95.13 | #237 |
| 3 | Gemini 2.5 Pro | 95.01 | #145 |
| 4 | Claude Sonnet 4.5 | 94.38 | #138 |
| 5 | GPT-5 | 94.28 | #91 |
| 6 | GLM-4.5V | 94.08 | #339 |
| 7 | Qwen 3 VL 32B Instruct | 94 | #276 |
| 8 | Claude Sonnet 4 (20250514) | 93.82 | #211 |
| 9 | Claude 3.7 Sonnet (20250219) | 93.51 | #196 |
| 10 | InternVL3-78B | 93.34 | #345 |
| 11 | GPT-4.1 | 92.55 | #240 |
| 12 | Qwen 3 VL 235B A22B Instruct | 90.22 | #264 |
| 13 | GPT-4o | 89.08 | #333 |
| 14 | Qwen 3 VL 30B A3B Instruct | 88.54 | #365 |
| 15 | InternVL3-38B | 87.94 | #395 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=st-bibench-foundational-spatial-grounding · How It Works · Data refreshed daily, snapshot 2026-10-11.