PrepBench - Code-to-Workflow Translation: leaderboard

Accuracy (%) with which LLM agents translate ground-truth preparation code into an executable GUI operator workflow on PrepBench (2026), a benchmark of natural-language data-preparation requests over imperfect real tables. Higher is better.

Metric: Translation accuracy (%): given the ground-truth preparation code, the agent must translate it into an executable GUI workflow with the same output, PrepBench natural-language data-preparation tasks, LLM agent at temperature 0.7 with Clarify (budgeted, answered by a DeepSeek-V3.2 user simulator), Profile, Code and Translate actions; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.1 Codex67.7#295
2Gemini 3 Flash52.3#175
3Kimi K2 (Thinking)46.1
4Grok Code Fast 145.1#703
5Claude Sonnet 4.544.8#261
6Qwen 3 235B A22B35#628
7GLM-4.734.6#380
8DeepSeek V3.227.8#419
9GPT-4o19.9#751
10Devstral 210.5#1010

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=prepbench-code-to-workflow-translation · How It Works · Data refreshed daily, snapshot 2026-11-21.