AgentHorizon
Harbor: pendingA benchmark of 1,373 computer-use tasks with paired instructions to evaluate agentic judges on long-horizon trajectory success verification across three OSs.
- Samples
- 1,373
- Released
- 2026
- Scores
- 10
HeRE
HeRE: A Novel Benchmark for Hebrew Relation Extraction via LLM-Guided Denoising of Knowledge Graph Alignments
HeRE is a 1.5M-instance silver dataset and gold-standard test sets for evaluating Hebrew Relation Extraction performance using LLMs.
How to run
ltb · here
$ npm i -g lettokenburn-cli$ ltb bench info here
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
12 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-5.4 | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Gemini 3 Flash Preview | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| google/gemma-4-31B-it | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Claude Haiku 4.5 | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| google/gemma-4-26B-A4B-it | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Gemma 3 27B | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Qwen3.5 35B A3B | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| DictaLM-3.0-24B | 0.9 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Aya-Expanse-32b | 0.8 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Gemma 2 9B IT | 0.8 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| XLM-R-XL | 0.8 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| NeoDictaBERT | 0.8 | 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
HeRE provides a massive silver-standard dataset of 1.5 million instances and two gold-standard sets of 500 human-annotated examples each for Hebrew Relation Extraction. It was constructed by denoising noisy alignments between Wikidata and Wikipedia texts using an NLI-based pipeline.
Scoring uses F1 score on the held-out test set. The best open-weights model, Gemma-4-31B-it, achieved an F1 of 0.844, demonstrating competitive performance against API models and validating the benchmark's utility for low-resource languages.