CITBench - Task Completion: leaderboard
CITBench (2026) evaluates LLMs on tabular data processing tasks of matching, cleaning, augmentation and transformation. Task completion is the share of runs whose output table is fully correct, averaged over the four offline difficulty tiers, in percent; higher is better.
Metric: Average task completion (%; share of repeated runs whose output table is fully correct, a Level-Accuracy of 1, on the offline single-turn tabular data processing tasks of CITBench, a balanced 50% slice of its 1,296 instances in 18 task types; temperature 1.0; unweighted mean of the Base, Complex-Rule, Multi-Input and Multi-Step tiers). Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 66.85 | #244 |
| 2 | Gemini 3 Flash | 60.12 | #159 |
| 3 | Gemini 3 Pro | 60.11 | #135 |
| 4 | GLM-4.7 | 57.62 | #365 |
| 5 | GPT-5.2 | 55.56 | #172 |
| 6 | Claude Haiku 4.5 | 50.43 | #568 |
| 7 | DeepSeek V3.2 | 49.17 | #403 |
| 8 | Qwen 3 235B A22B | 45.69 | #622 |
| 9 | Qwen 3 Coder Flash | 45.09 | #1624 |
| 10 | GPT-4o | 43.65 | #741 |
| 11 | Qwen 3 32B | 42.97 | #960 |
| 12 | Qwen 3 14B | 37.15 | #1235 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=citbench-task-completion · How It Works · Data refreshed daily, snapshot 2026-11-12.