ST-BiBench - Independent Parallel Manipulation: leaderboard
Metric: Success rate (%) over the 6 independent parallel manipulation tasks (place, rank and stack with both arms at once), each run for at least 100 domain-randomized episodes in which the model plans a sequence of seven parameterized bimanual action primitives executed through an API, averaged over the tasks it was scored on; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 78.5 | #240 |
| 2 | GPT-5 | 76.67 | #91 |
| 3 | Gemini 2.5 Pro | 71.33 | #145 |
| 4 | Claude 3.7 Sonnet (20250219) | 69 | #196 |
| 5 | Gemini 2.5 Flash | 67.17 | #237 |
| 6 | Claude Sonnet 4 (20250514) | 67 | #211 |
| 7 | Gemini 2.0 Flash | 62.83 | #331 |
| 8 | Qwen 3 VL 235B A22B Instruct | 58.67 | #264 |
| 9 | InternVL3-38B | 57.5 | #395 |
| 10 | InternVL3-78B | 56.33 | #345 |
| 11 | Qwen 3 VL 32B Instruct | 54.67 | #276 |
| 12 | Qwen 2.5 VL 32B Instruct | 52.67 | #443 |
| 13 | GPT-4o | 52.33 | #333 |
| 14 | InternVL2.5-78B | 47.83 | #421 |
| 15 | Qwen 2.5 VL 72B Instruct | 28.6 | #364 |
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-independent-parallel-manipulation · How It Works · Data refreshed daily, snapshot 2026-10-11.