DataClawEval: leaderboard

Metric: Rule-based task score (%; per task a weighted sum, usually 0.7 and 0.3, of the artifact score, checked row by row against live databases by task-specific deterministic scripts, and the process score for exploration, execution efficiency and self-verification, mean over all 100 tasks; one run per task, every model in the same Tencent CodeBuddy agent scaffold). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 16 models tracked.

Top models

#ModelScore
1CodeBuddy + GPT 5.574.9
2CodeBuddy + Claude Opus 4.874.3
3CodeBuddy + Claude Sonnet 573.8
4CodeBuddy + Gemini 3.1 Pro73.7
5CodeBuddy + Gemini 3.5 Flash73.3
6CodeBuddy + DeepSeek V4 Flash73
7CodeBuddy + MiniMax M371.8
8CodeBuddy + GLM 5.171.6
9CodeBuddy + DeepSeek V4 Pro70.6
10CodeBuddy + Kimi K2.669

Interactive version: theaggregate.ai/benchmark?slug=dataclaweval · How It Works · Data refreshed daily, snapshot 2026-09-29.