ARC-AGI-2
Harbor nativeARC-AGI-2: 更难的抽象推理
Verified100%Abstract Reasoning
- Samples
- 167
- Released
- 2025
- Scores
- 120
Event-STVG
Event-STVG: Spatio-Temporal Video Grounding for Long-Form Event Reasoning
A benchmark of 4,016 annotated event occurrences across 8 domains requiring joint spatio-temporal grounding and role inference in long-form videos.
How to run
ltb · event-stvg
$ npm i -g lettokenburn-cli$ ltb bench info event-stvg
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
10 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Gemini 3.7 Flash | 10.5 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Qwen3.8 Max | 4.0 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| GPT-5.6 Sol | 2.3 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| PTD | 1.7 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| GLM-5.3 Flash | 0.7 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Qwen3.7 Plus | 0.4 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| GLM-5V Turbo | 0.4 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| GLM-4.6V | 0.1 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| qwen3.5-9B | 0.1 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| MiniMax-M3 | 0.0 | N/A | Independent | medium | Oct 15, 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
Event-STVG comprises 4,016 annotated events spanning soccer and seven other domains, focusing on long-form reasoning, repeated-event disambiguation, and multi-role grounding. It evaluates models on their ability to localize specific event occurrences and their semantic participants in untrimmed videos.
Scoring uses mean video Intersection over Union (m-vIoU). State-of-the-art MLLMs struggle significantly, with the best model achieving only 5.4% m-vIoU at the 10-minute horizon, highlighting challenges in temporal context and role grounding.