MultiInpaintCOCO - 2img - az (English Prompt): leaderboard
MultiInpaintCOCO v1, 953 image-caption minimal pairs across 14 languages; az 2img bidirectional selection accuracy, English instruction prompt. Decoder-only checkpoint identities are bound by Appendix Table12; both answer orders are evaluated. Dual encoders, correlations, agreement and bootstrap uncertainty summaries are separate analyses. Reported metric: Accuracy (fraction).
Metric: Accuracy (fraction). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 0.58 | #419 |
| 2 | Gemma 3 12B (IT) | 0.5 | #672 |
| 3 | Qwen 2.5 VL 7B Instruct | 0.41 | #650 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=multiinpaintcoco-2img-az-english-prompt · How It Works · Data refreshed daily, snapshot 2026-11-05.