SADU - Retrieval: leaderboard

SADU (2026) tests vision-language models on counting and retrieval questions about behavioral, structural and entity-relationship software architecture diagrams. This board is the exact-match rate in percent on the retrieval questions, where the predicted element set must equal the gold set; higher is better.

Metric: Exact-match rate (%) on the retrieval questions of SADU, the predicted element set equal to the gold set (154 software architecture diagrams (behavioral, structural and ER) with 2,431 counting and retrieval questions, temperature 0); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Flash Lite (Preview)68#417
2Gemini 2.5 Flash63#463
3Claude Sonnet 4.558#261
4Gemini 3 Flash (Preview)58#152
5GPT-5 Nano57#906
6Gemini 2.5 Flash Lite53#964
7Qwen 2.5 VL 32B53#941
8Claude Haiku 4.549#583
9Qwen 2.5 VL 7B46#1862
10Qwen 2.5 VL 3B26#2639
11GPT-4o Mini18#1667

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=sadu-retrieval · How It Works · Data refreshed daily, snapshot 2026-11-21.