FlagEval EmbodiedVerse - ERQA: leaderboard
BAAI FlagEval’s EmbodiedVerse suite, ERQA component: embodied reasoning question answering over robot-perspective 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 | 59.25 | #98 |
| 2 | Gemini 2.5 Pro | 57 | #152 |
| 3 | GPT-4.1 | 54 | #248 |
| 4 | Qwen 3 VL 235B A22B Instruct | 51.75 | #273 |
| 5 | Qwen 3 VL 235B A22B (Thinking) | 51 | #238 (Qwen 3 VL 235B A22B) |
| 6 | Qwen 3 VL 30B A3B Instruct | 46.75 | #384 |
| 7 | Qwen 3 VL 30B A3B (Thinking) | 45.25 | #351 (Qwen 3 VL 30B A3B) |
| 8 | Claude Sonnet 4.5 | 44.25 | #142 |
| 9 | Qwen 2.5 VL 32B Instruct | 42 | #453 |
| 10 | InternVL3-78B | 41.5 | #352 |
| 11 | Qwen 2.5 VL 72B Instruct | 40.5 | #377 |
| 12 | O4 Mini | 39.75 | #180 |
| 13 | Mistral Small 3.2 | 39 | #605 |
| 14 | InternVL3-8B | 37 | #620 |
| 15 | Qwen 2.5 VL 7B Instruct | 36.25 | #654 |
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-erqa · How It Works · Data refreshed daily, snapshot 2026-10-20.