AtelierEval - Prompt Quality (Novice Prompting): leaderboard
AtelierEval (Luo et al., 2026) tests models as text-to-image prompters over 360 expert-written tasks. A GPT-5.4 judge calibrated with expert-scored example prompts rates each prompt from 1 to 5 on instructional clarity, creative elaboration, terminology and intent formalization; the board is the mean rating (higher is better). The model receives a direct natural-language request to write the prompt.
Metric: Judge score (1-5; mean of four prompt-level Likert dimensions from a memory-augmented GPT-5.4 judge calibrated on expert-scored exemplars, mean of the three task categories; direct natural-language prompting). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 4.05 | #172 |
| 2 | Gemini 3 Pro (Preview) | 3.81 | #111 |
| 3 | Qwen 3 VL 235B A22B | 3.65 | #477 |
| 4 | GPT-4.1 | 3.46 | #492 |
| 5 | Claude Sonnet 4.5 | 3.45 | #244 |
| 6 | GPT-4.1 Nano | 3.27 | #1992 |
| 7 | Gemini 2.0 Flash | 2.96 | #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-quality-novice-prompting · How It Works · Data refreshed daily, snapshot 2026-11-12.