3DVerifierBench
Harbor adapterDVerifierBench Grounding Generative Verification in 3D ContentShuanglin Li, Jiaying Lin, Shuquan Ye.
- Samples
- 2,581
- Released
- 2026
- Scores
- 10
OmniCompass
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
A benchmark for omni-modal video grounding and understanding with 820 videos (2,460 instances) across VL and AV tracks, covering temporal/spatial grounding and QA.
How to run
ltb · omnicompass
$ npm i -g lettokenburn-cli$ ltb bench info omnicompass
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
13 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Gemini 3.8 Flash | 50.6 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Seed-2.1-Pro-260915 | 47.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Seed-2.1-Turbo-260628 | 46.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Seed-2.0-Pro-260215 | 46.4 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3.8 Max | 46.2 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Gemini 3.1 Pro | 45.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Kimi K3 | 45.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-6 Astra | 45.2 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5.6 Sol | 44.5 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3.8 Flash | 43.6 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5.5 | 41.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5.6 Luna | 40.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5.6-terra | 40.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
OmniCompass evaluates MLLMs on temporal grounding, spatial grounding, and question answering using paired text and text+image specifications. It includes 820 videos across short, medium, and long durations, with a subset requiring audio-visual evidence.
Scoring uses Cross-Specification Consistency Scores to measure accuracy and consistency. Evaluations of 50 models show proprietary models outperform open-source ones, with top scores of 42.8% and 39.6% respectively.