EXPLORE-Bench - Object Coverage: leaderboard
Metric: Object-level coverage (0-100): soft Sentence-BERT matching of the objects named in the predicted final-scene description to the annotated objects, over all 1,157 instances; the model sees an initial egocentric scene image and a sequence of atomic actions (11 to 694) from real Ego4D, Ego-Exo4D and self-recorded videos and describes the final scene; greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B (Thinking) | 62.7 | |
| 2 | Gemini 3 Pro | 60.94 | #77 |
| 3 | Qwen 3 VL 8B Instruct | 60.63 | #401 |
| 4 | Step3 VL 10B | 60.61 | #465 |
| 5 | Gemini 3 Flash | 59.72 | #93 |
| 6 | GPT-5.2 Instant | 59.69 | #205 |
| 7 | Qwen 2.5 VL 7B Instruct | 51.67 | #643 |
| 8 | Qwen 2 VL 7B Instruct | 46.62 | #816 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=explore-bench-object-coverage · How It Works · Data refreshed daily, snapshot 2026-10-11.