OmniGen2: benchmark results

Provider: Other.

Aggregate Rating 1435 ± 48, rank #4394 of 10916 rated models, from 155 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
UniG2U-Bench (Generate-then-Answer) - Physics Reasoning41.5Accuracy (%) on the 200 mechanics and optics items, mean of 100
UniG2U-Bench (Generate-then-Answer) - Real-world Applications41.25Accuracy (%) on the 200 real-world application items (attent100
UniSAFE - Image Composition ASR26.1Attack success rate (%): share of UniSAFE's unsafe-intent pr100
UniSAFE - Multi-Turn Editing ASR17.4Attack success rate (%): share of UniSAFE's unsafe-intent pr100
KVBench (Brief Caption) - Math (Chinese)36.31Checklist score (%) on the math prompts in Chinese, each giv92.3
UniG2U-Bench (Direct) - Physics Reasoning41Accuracy (%) on the 200 mechanics and optics items, mean of 92.3
UniG2U-Bench (Generate-then-Answer) - Spatial Intelligence26.4Accuracy (%) on the 500 MMSI-Bench spatial items (multi-step90
UniG2U-Bench (Direct) - Real-world Applications41.25Accuracy (%) on the 200 real-world application items (attent84.6
Unison - Internal Consistency74.5Unified score (%) on the internal consistency dimension of U78.6
UniSAFE - Text-to-Image ASR55.8Attack success rate (%): share of UniSAFE's unsafe-intent pr76.9
UniG2U-Bench (Direct) - Chart Reasoning86Accuracy (%) on the 100 ChartQA items, from UniG2U-Bench, an75
UniG2U-Bench (Generate-then-Answer) - Chart Reasoning79Accuracy (%) on the 100 ChartQA items, from UniG2U-Bench, Ge75

Interactive version: theaggregate.ai/model?slug=omnigen2 · How It Works · Data refreshed daily, snapshot 2026-11-08.