Ovis2.5-9B: benchmark results
Provider: Alibaba. Released 2025-08-15. Access: Open.
Aggregate Rating 1574 ± 16, rank #1335 of 10916 rated models, from 115 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FairLens | 93.8 | Soundness (%; share of closed-ended answers matching the sou | 100 |
| FairLens - Healthcare | 98.5 | Soundness (%; healthcare-domain closed-ended questions, 8 it | 100 |
| FairLens - Hiring | 98.5 | Soundness (%; hiring-domain closed-ended questions, 21 items | 100 |
| HallusionBench | 65.3 | Overall Score | 98.7 |
| MathVista | 82.1 | Accuracy (%) | 97.8 |
| SenseBench - Single-Image Description | 4.75 | Judge Score Sum (0-6; Prometheus completeness, correctness a | 96.2 |
| MMStar | 72.5 | Accuracy (%) | 93.2 |
| MIOH - Counting | 32.7 | Accuracy (%; counting questions about how many objects there | 92.6 |
| SenseBench - Paired-Image Description | 4.21 | Judge Score Sum (0-6; Prometheus completeness, correctness a | 90.5 |
| AI2D | 87.4 | Accuracy (%) | 89.4 |
| TableVision - Compositional Arithmetic | 66.4 | Exact-match accuracy (%) on L3 analysis, compositional arith | 88.9 |
| TableVision - Counting | 58.7 | Exact-match accuracy (%) on L2 reasoning, counting cells (92 | 88.9 |
Interactive version: theaggregate.ai/model?slug=ovis2-5-9b · How It Works · Data refreshed daily, snapshot 2026-11-08.