FlagEval EmbodiedVerse - EmbSpatial-Bench: leaderboard
BAAI FlagEval’s EmbodiedVerse suite, EmbSpatial-Bench component: spatial relations between objects as seen from an embodied agent’s viewpoint. Scored on FlagEval’s common scale so the thirteen component boards read against each other.
Metric: Score. Source: flageval.baai.ac.cn. Saturation forecast: Around January 2027. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B (Thinking) | 83.49 | #238 (Qwen 3 VL 235B A22B) |
| 2 | Qwen 3 VL 235B A22B Instruct | 82.66 | #273 |
| 3 | GPT-5 | 82.64 | #98 |
| 4 | Qwen 3 VL 30B A3B (Thinking) | 79.45 | #351 (Qwen 3 VL 30B A3B) |
| 5 | Gemini 2.5 Pro | 78.68 | #152 |
| 6 | Qwen 3 VL 30B A3B Instruct | 76.29 | #384 |
| 7 | GPT-4.1 | 76.04 | #248 |
| 8 | InternVL3-78B | 74.59 | #352 |
| 9 | Qwen 2.5 VL 32B Instruct | 74.45 | #453 |
| 10 | Qwen 2.5 VL 72B Instruct | 73.08 | #377 |
| 11 | InternVL3-8B | 71.98 | #620 |
| 12 | Qwen 2.5 VL 7B Instruct | 70.33 | #654 |
| 13 | Claude Sonnet 4.5 | 69.59 | #142 |
| 14 | Mistral Small 3.2 | 66.62 | #605 |
| 15 | O4 Mini | 55.71 | #180 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=flageval-embodiedverse-embspatial-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.