OmniLMM-12B: benchmark results
A multimodal checkpoint combining EVA02-5B and Zephyr-7B-beta through a perceiver resampler, trained on multimodal data and aligned with multimodal RLHF. The 7B language component alone does not describe the complete model. Provider: OpenBMB. Released 2024-02-01. Access: Open.
Aggregate Rating 1329 ± 30, rank #8179 of 10916 rated models, from 410 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM MMT-Bench - Point Tracking | 90 | Score (%) | 98.1 |
| OpenVLM MMT-Bench - Other Biological Attributes | 90 | Score (%) | 97.8 |
| OpenVLM MMT-Bench - Temporal Anticipation | 85 | Score (%) | 97.8 |
| OpenVLM MMT-Bench - Spot The Diff | 95 | Score (%) | 97.3 |
| OpenVLM MMT-Bench - Lesion Grading | 80 | Score (%) | 94.4 |
| OpenVLM MMT-Bench - Attribute Hallucination | 85 | Score (%) | 92.5 |
| OpenVLM MMT-Bench - Food Recognition | 100 | Score (%) | 92.5 |
| OpenVLM MMT-Bench - Order Hallucination | 55 | Score (%) | 92.2 |
| OpenVLM MME - Celebrity | 169.7 | Score | 91.9 |
| OpenVLM MME - Artwork | 156.2 | Score | 90.2 |
| OpenVLM MMT-Bench - MedU | 79 | Score (%) | 88.8 |
| OpenVLM MMT-Bench - Disaster Recognition | 95 | Score (%) | 88.6 |
Interactive version: theaggregate.ai/model?slug=omnilmm-12b · How It Works · Data refreshed daily, snapshot 2026-11-08.