SciIR-Bench - Scientific Process: leaderboard
SciIR-Bench (2026) tests text-to-image models on 800 scientific illustration prompts drawn from open-access publications; a Gemini 3 Pro reviewer answers atomic yes or no checks, and an image counts for a track only if it passes every check of that track. This board is accuracy in percent on the Scientific Process track; higher is better.
Metric: Accuracy (%) on the Scientific Process (index) track; a gemini-3-pro-preview reviewer answers the atomic binary checks generated for each prompt with visual evidence retrieval; a sample counts only if it passes every check of the track; averaged over intrinsic-reasoning (abstract prompt) and instruction-following (dense scientific reasoning chain prompt) samples; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Nano Banana Pro | 97 | |
| 2 | GPT-Image-1 | 72 | |
| 3 | Seedream 4.5 | 60 | |
| 4 | Qwen-Image-2512 | 37 | |
| 5 | FLUX-Kontext-Max | 28 | |
| 6 | Show-o2-7B | 28 | |
| 7 | HiDream-I1-Full | 16 | |
| 8 | FLUX.1-dev | 12 | |
| 9 | Stable Diffusion 3.5 Large | 6 | |
| 10 | BAGEL-7B-MoT | 2 | #5470 |
Interactive version: theaggregate.ai/benchmark?slug=sciir-bench-scientific-process · How It Works · Data refreshed daily, snapshot 2026-11-13.