AtelierEval - Prompt Constraint Coverage (Skilled Prompting): leaderboard
AtelierEval (Luo et al., 2026) tests models as text-to-image prompters: given a user intent (an open-ended request, a structured specification or a target image), the model writes one image-generation prompt. The board is the share of the expert constraint checklist of each task that the prompt itself specifies, checked by a GPT-5.4 judge and averaged over 360 tasks in three categories, in percent (higher is better). The model receives a structured system prompt asking for chain-of-thought analysis and self-verification.
Metric: Prompt checklist coverage (%; share of each task's expert constraint checklist that the written prompt specifies, checked by a GPT-5.4 AtelierJudge, mean of the open-ended, constrained and imitation categories, 120 tasks each; structured prompting with chain-of-thought and self-verification). Source: arxiv.org. Saturation forecast: Around July 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 65.5 | #111 |
| 2 | GPT-5.2 | 63.7 | #172 |
| 3 | Claude Sonnet 4.5 | 60.1 | #244 |
| 4 | Qwen 3 VL 235B A22B | 58.8 | #477 |
| 5 | GPT-4.1 | 56.1 | #492 |
| 6 | GPT-4.1 Nano | 53.2 | #1992 |
| 7 | Gemini 2.0 Flash | 48.2 | #702 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ateliereval-prompt-constraint-coverage-skilled-prompting · How It Works · Data refreshed daily, snapshot 2026-11-12.