AtelierEval - Prompt Constraint Coverage - Open-Ended Creation (Skilled Prompting): leaderboard
The 120 open-ended creation tasks of AtelierEval (Luo et al., 2026): an abstract, narrative request with implicit style and audience cues that the model must turn into one text-to-image prompt. Scored as the share of the constraint checklist of the task that the prompt specifies, checked by a GPT-5.4 judge, 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; 120 open-ended creation tasks: abstract, narrative requests to turn into an executable prompt; structured prompting with chain-of-thought and self-verification). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 92.5 | #111 |
| 2 | GPT-5.2 | 90.8 | #172 |
| 3 | Qwen 3 VL 235B A22B | 88.4 | #477 |
| 4 | Claude Sonnet 4.5 | 88.2 | #244 |
| 5 | GPT-4.1 | 87 | #492 |
| 6 | GPT-4.1 Nano | 83.8 | #1992 |
| 7 | Gemini 2.0 Flash | 83.1 | #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-open-ended-creation-skilled-prompting · How It Works · Data refreshed daily, snapshot 2026-11-12.