OMD-Bench (Synthetic) - Abstention Calibration Error: leaderboard
Metric: Abstention calibration error (%): the mean over corruption levels k=0 to 3 of the absolute gap between the model's abstention rate and the human-annotated rate (near 0 up to k=2, near 98 at k=3), on the synthetic split (generated video and audio with human-written text) of OMD-Bench (255 knowledge questions about 27 anchors, objects or events shown at once in video, audio and text; five options including an explicit abstention; unparseable answers are excluded), zero-shot prompting; lower is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 18.9 | #362 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=omd-bench-synthetic-abstention-calibration-error · How It Works · Data refreshed daily, snapshot 2026-10-11.