FlagEval EmbodiedVerse - VSI-Bench (tiny): leaderboard
BAAI FlagEval’s EmbodiedVerse suite, VSI-Bench (tiny) component: the tiny split of VSI-Bench, video-based spatial intelligence over walked-through scenes. 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 November 2026. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 57.53 | #98 |
| 2 | Qwen 3 VL 235B A22B (Thinking) | 54.74 | #238 (Qwen 3 VL 235B A22B) |
| 3 | Gemini 2.5 Pro | 49.13 | #152 |
| 4 | Qwen 3 VL 235B A22B Instruct | 48.67 | #273 |
| 5 | Qwen 3 VL 30B A3B Instruct | 47.49 | #384 |
| 6 | Qwen 3 VL 30B A3B (Thinking) | 45.92 | #351 (Qwen 3 VL 30B A3B) |
| 7 | GPT-4.1 | 45.36 | #248 |
| 8 | Claude Sonnet 4.5 | 45.28 | #142 |
| 9 | InternVL3-78B | 45.03 | #352 |
| 10 | Qwen 2.5 VL 72B Instruct | 39.62 | #377 |
| 11 | Qwen 2.5 VL 32B Instruct | 39.13 | #453 |
| 12 | InternVL3-8B | 38.09 | #620 |
| 13 | O4 Mini | 33.47 | #180 |
| 14 | Qwen 2.5 VL 7B Instruct | 17.17 | #654 |
| 15 | Phi-4 Multimodal Instruct | 16.22 | #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-vsi-bench-tiny · How It Works · Data refreshed daily, snapshot 2026-10-20.