CITBench - Base: leaderboard

Metric: Average performance (%; mean Level-Accuracy with partial credit, each of repeated runs scored 0 to 1, on the offline single-turn tabular data processing tasks of CITBench, a balanced 50% slice of its 1,296 instances in 18 matching, cleaning, augmentation and transformation task types; temperature 1.0, 8,192 output tokens; Base tier: one input table under simple rules). Source: arxiv.org. Saturation forecast: Around December 2026. 12 models tracked.

Top models

#ModelScore
1Gemini 3 Flash92.26
2Claude Sonnet 4.591.35
3Gemini 3 Pro86.39
4GPT-5.285.79
5GLM-4.785.57
6GPT-4o80.85
7DeepSeek V3.277.4
8Qwen 3 235B A22B76.03
9Qwen 3 Coder Flash74.66
10Claude Haiku 4.574.23
11Qwen 3 32B73.22
12Qwen 3 14B69.34

Interactive version: theaggregate.ai/benchmark?slug=citbench-base · How It Works · Data refreshed daily, snapshot 2026-09-26.