Human Creativity Benchmark - Desktop Apps (Refinement) - Prompt Adherence: leaderboard
Metric: Mean expert Likert rating (1-5) of faithfulness to the prompt; desktop-app interfaces written as code at the Refinement stage, rated by professional creatives. Source: arxiv.org. Saturation forecast: Around December 2026. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.3 Codex | 4.03 |
| 2 | Gemini 3.1 Pro (Preview) | 3.74 |
| 3 | Qwen 3.5 397B A17B | 3.74 |
| 4 | Claude Opus 4.6 | 3.69 |
Interactive version: theaggregate.ai/benchmark?slug=human-creativity-benchmark-desktop-apps-refinement-prompt-adherence · How It Works · Data refreshed daily, snapshot 2026-09-26.