R2-Omni - OmniDCBench SODA-M (Uncompressed): leaderboard
R2-Omni v1 Table1, original uncompressed Omni-LLMs with native audio-visual pipelines and greedy decoding. OmniDCBench SODA-M measures long-form audio-visual event description; event-error rate is lower-is-better. Compression methods and normalized relative-quality scores are kept outside the original-model baseline. No unsupported finite upper bound is assigned to EER or SODA metrics. Reported metric: OmniDCBench SODA-M.
Metric: OmniDCBench SODA-M. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen2.5-Omni-7B | 7.32 | #633 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=r2-omni-omnidcbench-soda-m-uncompressed · How It Works · Data refreshed daily, snapshot 2026-11-05.