TECCI - Instruction Rich Challenge Set: leaderboard
The manually written challenge set of TECCI (Agrawal et al., 2026): hard, often creative edit instructions for images picked for rare object and attribute combinations. Overall human success rate on 265 sampled edits, in percent, requiring a mean rating of 4.5 of 5 on all three criteria; higher is better.
Metric: Overall human success rate (%; five trained raters score each edit 1-5 and a criterion succeeds when the mean is at least 4.5 on all three criteria; 265 sampled pairs with manually written challenging instructions). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Nano Banana Pro | 19.7 | |
| 2 | Nano Banana 2 (Gemini 3.1 Flash Image Preview) | 17.7 | |
| 3 | Grok Imagine Pro | 14.7 | |
| 4 | Seedream 5.0 Lite | 10.3 | |
| 5 | GPT Image 1.5 | 4.9 |
Interactive version: theaggregate.ai/benchmark?slug=tecci-instruction-rich-challenge-set · How It Works · Data refreshed daily, snapshot 2026-11-12.