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
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Non-reasoning) | 13.99 |
| 2 | Claude Opus 4.6 | 12.24 |
| 3 | Claude Opus 4.5 | 11.54 |
| 4 | GLM-5 (Non-reasoning) | 10.49 |
| 5 | Kimi K2.6 (Non-reasoning) | 9.44 |
| 6 | Qwen 3.6 27B (Non-reasoning) | 8.42 |
| 7 | Gemma 4 31B (IT) (Non-reasoning) | 8.04 |
| 8 | Qwen 3.6 Max | 7.34 |
| 9 | Qwen 3.6 35B A3B (Non-reasoning) | 7.02 |
| 10 | DeepSeek V4 Flash (Non-reasoning) | 4.9 |
| 11 | Llama 4 Scout | 3.87 |
Interactive version: theaggregate.ai/benchmark?slug=cdr-bench-order-sensitive-mapper-recipes · How It Works · Data refreshed daily, snapshot 2026-09-29.