CDR-Bench - Order-Sensitive Mapper Recipes: leaderboard

Metric: Recipe success RS@3 (%) on the order-sensitive mapper-recipe tasks of CDR-Bench (3,462 data-refinement tasks over web, LaTeX, RAG-preparation and privacy text with 29 Data-Juicer operators and deterministic reference outputs): a task counts when the model's refined text or KEEP/DROP judgment exactly matches the reference within 3 attempts; non-thinking mode, baseline prompt; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 11 models tracked.

Top models

#ModelScore
1GPT-5.4 (Non-reasoning)13.99
2Claude Opus 4.612.24
3Claude Opus 4.511.54
4GLM-5 (Non-reasoning)10.49
5Kimi K2.6 (Non-reasoning)9.44
6Qwen 3.6 27B (Non-reasoning)8.42
7Gemma 4 31B (IT) (Non-reasoning)8.04
8Qwen 3.6 Max7.34
9Qwen 3.6 35B A3B (Non-reasoning)7.02
10DeepSeek V4 Flash (Non-reasoning)4.9
11Llama 4 Scout3.87

Interactive version: theaggregate.ai/benchmark?slug=cdr-bench-order-sensitive-mapper-recipes · How It Works · Data refreshed daily, snapshot 2026-09-29.