PiSAR - App Store Reviews: leaderboard
The app-store slice of PiSAR (Bissa, Vyas and Jain, AprioriLabs, 2026): for 542 held-out public app reviews paired with the app's screenshots and a matched persona, the model writes the reviewer's rationale. Mean cosine similarity of text-embedding-3-small embeddings of the predicted and the real rationale (-1 to 1), higher is better.
Metric: Semantic similarity (mean cosine similarity between OpenAI text-embedding-3-small embeddings of the predicted and the gold rationale; 542 held-out app-store reviews paired with the app's screenshots). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 0.5 | #48 |
| 2 | Claude Opus 4.7 | 0.49 | #75 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pisar-app-store-reviews · How It Works · Data refreshed daily, snapshot 2026-11-12.