DataVista
Harbor: pendingA benchmark with 961 real-world data videos and 6,775 multiple-choice questions evaluating MLLMs on data perception, temporal reasoning, and narrative understanding.
- Samples
- 6,775
- Released
- 2026
- Scores
- 13
Knossos
A benchmark of 19,200 synthetic structural diagrams across six domains for evaluating complete diagram-to-graph topology extraction.
How to run
ltb · knossos
$ npm i -g lettokenburn-cli$ ltb bench info knossos
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 | |||
|---|---|---|---|---|---|---|
| Ariadne | 91.6 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| GPT-5.6 | 81.1 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| Claude Opus 5 | 70.4 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| Qwen3-VL-32B | 54.7 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| TopoAgent | 46.5 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| Qwen3-VL-8B-Instruct | 46.1 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| Speculative Verdict | 43.9 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| GPT-4o | 41.2 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| InternVL3.5-8B-Instruct | 40.7 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| MiMo-VL-7B-SFT | 40.7 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| Keye-VL-1.5-8B | 39.0 | N/A | Independent | medium | Oct 3, 2026 | arxiv.org |
| Ovis2.5-9B | 38.3 | N/A | Independent | medium | Oct 3, 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
Knossos contains 19,200 synthetic diagrams from six domains (e.g., Food Web, Circuit) with exact annotations for nodes, typed edges, and connector geometry. It enables supervised training and evaluation of vision-language models on complete topology extraction tasks.
Evaluation uses Edge F1 to measure the accuracy of extracted graph structures. Models trained on Knossos show substantial improvements in topology extraction, with the Ariadne framework achieving the highest average Edge F1 among evaluated methods.