FlagEval EmbodiedVerse - Where2Place: leaderboard
BAAI FlagEval’s EmbodiedVerse suite, Where2Place component: placement grounding, asking where in the scene an object should go. 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 February 2027. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B Instruct | 58.35 | #273 |
| 2 | Qwen 3 VL 235B A22B (Thinking) | 52.61 | #238 (Qwen 3 VL 235B A22B) |
| 3 | Qwen 3 VL 30B A3B (Thinking) | 38.82 | #351 (Qwen 3 VL 30B A3B) |
| 4 | GPT-5 | 38.5 | #98 |
| 5 | Qwen 2.5 VL 72B Instruct | 38.35 | #377 |
| 6 | GPT-4.1 | 36.67 | #248 |
| 7 | Gemini 2.5 Pro | 36.27 | #152 |
| 8 | InternVL3-78B | 24.12 | #352 |
| 9 | Qwen 2.5 VL 32B Instruct | 21.63 | #453 |
| 10 | Mistral Small 3.2 | 16.66 | #605 |
| 11 | Claude Sonnet 4.5 | 15.8 | #142 |
| 12 | InternVL3-8B | 14.6 | #620 |
| 13 | Qwen 3 VL 30B A3B Instruct | 13.16 | #384 |
| 14 | O4 Mini | 9.97 | #180 |
| 15 | Phi-4 Multimodal Instruct | 5.71 | #906 |
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-where2place · How It Works · Data refreshed daily, snapshot 2026-10-20.