Is GenEval 2 still meaningful?

Known problems

• The 35.8% figure comes from one secondary overview of the paper.^{[2][2] alphaXiv overview of GenEval 2 Single source}
• Its scale differs from GenEval's, so the drop from 96.7% marks a reset, not a model regression.^{[2][2] alphaXiv overview of GenEval 2 Single source}

What to look at instead

Arena and Artificial Analysis agree on the image leader: GPT-Image-2.5 Sunburst tops both boards.^{[3][3] Arena, text-to-image leaderboard Primary[4][4] Artificial Analysis, text-to-image leaderboard Primary}

• Chatbot Arena (LMArena) Status:Contested Partly. It still shows crowd preference, but private testing and unequal data access distort the ranking.

Where it appears in the report

Chapter 7, Multimodal, voice and embodied

• Figure 1. Tests Beyond Text Fill Up in Months, Too
• Table 1. Each modality has one number that tells its story
• Figure 4. A harder test drops the best image score from 97% to 36%
• Table 6. Automatic metrics and learned scorers for image, video and music generation, read November 2026 in “Go deeper: Automated generation metrics”
• Figure 11. Robotics, GUI agents and deepfake detection have the weakest independent checks