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
HyperBrowseComp
HyperBrowseComp: A Multilingual and Multimodal Stress Test for Web-Browsing Agents
HyperBrowseComp HyperBrowseComp is a multilingual and multimodal stress test for web-browsing agents.
How to run
ltb · hyperbrowsecomp
$ npm i -g lettokenburn-cli$ ltb bench info hyperbrowsecomp
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
6 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Gemini 3.7 Flash | 31.7 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Gemini 3.1 Pro | 25.5 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| GPT-5.6 Sol | 19.1 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Gemini 3.7 Flash | 17.7 | owl | Independent | medium | Oct 2, 2026 | arxiv.org |
| GPT-5.6-terra | 15.8 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| GPT-5.6 Luna | 11.6 | N/A | Independent | medium | Oct 2, 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
HyperBrowseComp
HyperBrowseComp is a multilingual and multimodal stress test for web-browsing agents. The paper's filtered benchmark contains 423 manually authored and human-validated questions across 13 languages. Each question has a short, publicly verifiable answer, but finding it can require following several clues across obscure webpages, videos, scanned documents, images, or maps. The benchmark tests whether agents can persistently find and connect evidence on the open…