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
PlotGround-1k
PlotGround: Grounding Plot Digitization in Real Scientific Figures and Their Source Data
A human-verified benchmark of 1,119 quantitative questions from 1,066 bioRxiv preprints measuring plot digitization accuracy using real figures and source data.
How to run
ltb · plotground-1k
$ npm i -g lettokenburn-cli$ ltb bench info plotground-1k
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
16 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-5.6 Sol | 73.8 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| GPT-5.6 Luna | 69.7 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| GPT-5.6-terra | 66.6 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Gemini 3.7 Flash | 65.7 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Gemini 3.8 Flash | 65.5 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Claude Opus 5 | 60.2 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Qwen3.8 Max | 57.7 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Claude Sonnet 5 | 55.6 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Gemini 3.1 Pro | 49.9 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Qwen3.8 Flash | 42.6 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| GLM-5.3 Flash | 39.8 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Kimi K3 | 38.2 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| MiniMax-M3 | 33.8 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Inkling | 32.0 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| DeepSeek V4.1 Flash | 29.3 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| DeepSeek V4 FV exp | 17.0 | N/A | Independent | medium | Oct 5, 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
PlotGround-1k consists of 1,119 questions derived from real scientific figures in bioRxiv preprints, grounded by author-released source data. It uses an automated pipeline to map figures to source tables and generate quantitative questions.
Scoring uses relative-error tolerance (±5% or ±2%). The best model achieved 87.5% accuracy at ±5% tolerance, with significant drops at stricter tolerances, highlighting a precision gap in visual numerical recovery.