AgentDoxx
Harbor: pendingAn evaluation suite of 822 synthetic interview transcripts grounded in public info to measure LLM re-identification risks via web search.
- Samples
- 822
- Released
- 2026
- Scores
- 9
NextAct
NextAct: Personalized Action Anticipation Benchmark
A 1,500-point benchmark for open-vocabulary K-step sequence prediction of personal behavior from egocentric video, combining NextMe-800 and EgoLife datasets.
How to run
ltb · nextact
$ npm i -g lettokenburn-cli$ ltb bench info nextact
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 | |||
|---|---|---|---|---|---|---|
| Qwen3.5 4B | 0.3 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Gemma4-12B-it | 0.3 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Claude Opus 4.6 | 0.3 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| DeepSeek V4 Flash | 0.3 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| GPT-5.6 Sol | 0.3 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Gemini 3.1 Pro | 0.3 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Qwen3.5 Flash | 0.9 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| GPT-5.6 Luna | 0.9 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Gemini 3.7 Flash | 0.9 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Claude Haiku 4.5 | 1.0 | N/A | Independent | medium | Oct 1, 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
NextAct combines 1,500 evaluation points from the NextMe-800 dataset (800 hours of single-user egocentric video) and the multi-person EgoLife dataset. It targets personalized action anticipation across five hierarchical abstraction levels.
Scoring uses an embedding-based soft edit distance to handle open-vocabulary predictions, validated against human rankings. The benchmark evaluates how well models anticipate future behavior across different prediction horizons and abstraction levels.