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
MMTA
Multimodal Temporal Analysis
MMTA benchmarks multimodal temporal analysis across 9 domains with 3,524 tri-modal (time series, text, vision) samples for classification and forecasting tasks.
How to run
ltb · mmta
$ npm i -g lettokenburn-cli$ ltb bench info mmta
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
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Gemini 3 Flash Preview | 0.4 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5-mini | 0.4 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3-Omni-30B-A3B-Inst. | 0.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3-VL-8B-Instruct | 0.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| InternVL3.5-8B-Instruct | 0.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| InternVL3.5-4B | 0.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3-VL-30B-A3B-Inst. | 0.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen2.5-Omni-7B | 0.3 | N/A | Independent | medium | Oct 13, 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
MMTA unifies nine real-world domains (e.g., robotics, healthcare) into a shared schema with video+TS or image+TS layouts. It standardizes heterogeneous sources with domain-specific prompts and provenance documentation.
The benchmark uses success-weighted metrics penalizing wrong answers and failed generations. Evaluations reveal that serializing time series as text creates long token sequences and weakens visual-temporal alignment in MLLMs.