Meta-Router
Harbor: pendingIn language tasks requiring extensive human-model interaction, the inference cost of large language models (LLMs) can be substantial.
Incomplete29%Knowledge & Reasoning
- Samples
- N/A
- Released
- 2026
- Scores
- 0
Meta-Video
Meta-Video: Efficient Standardized Evaluation for Video Understanding
A balanced 1,000-question benchmark over 791 videos for efficient video understanding evaluation, preserving model rankings with 5x speedup.
How to run
ltb · meta-video
$ npm i -g lettokenburn-cli$ ltb bench info meta-video
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
5 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Qwen3-VL-8B-Instruct | 58.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3-VL-4B | 55.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| InternVL3.5-8B-Instruct | 52.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| InternVL3.5-4B | 44.6 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen2.5-VL-7B | 40.6 | 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
Meta-Video curates 1,000 QA pairs from 791 videos across diverse tasks, filtering out questions solvable without visual input to ensure discriminability. It establishes a standardized evaluation protocol by isolating configuration choices like frame sampling and input formatting that cause significant score variance.
The benchmark preserves model rankings of source suites while requiring one-fifth of the inference compute for full frame-count sweeps. Reference scores are released for five models, demonstrating that unexamined evaluation defaults can shift accuracy by up to 19.5 percentage points.