Elliot Caption Quality - Mean of 5: leaderboard
Elliot Caption Quality - Mean of 5. 100 stratified tiles;blind ClaudeOpus5.5 judge sees image,fact sheet,four randomized captions. Mean over tiles; source Table5. Input conditions retained on exact run labels. Metric: Judge score (0–10). Primary published table; source configurations remain distinct.
Metric: Judge score (0–10). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6.1 Sol (Elliot Run: Fact sheet) | 8.25 | |
| 2 | GPT-6.1 Sol (Elliot Run: Fact sheet and image) | 8.24 | |
| 3 | GPT-6 Luna (Elliot Run: Fact sheet) | 7.26 |
Interactive version: theaggregate.ai/benchmark?slug=elliot-caption-quality-mean-of-5 · How It Works · Data refreshed daily, snapshot 2026-10-16.