PrepBench - GUI Workflow: leaderboard

End-to-end accuracy (%) of LLM agents that turn an ambiguous data-preparation request into an executable GUI operator workflow on PrepBench (2026), a benchmark of natural-language data-preparation requests over imperfect real tables. Agents may ask a simulated user clarification questions; higher is better.

Metric: End-to-end GUI workflow accuracy (%): from the original ambiguous request, the agent must produce an executable operator workflow whose output table matches the reference, 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 Codex34.6#295
2Kimi K2 (Thinking)30.1
3Claude Sonnet 4.524.5#261
4Gemini 3 Flash22.2#175
5GLM-4.719.9#380
6Qwen 3 235B A22B19.3#628
7DeepSeek V3.215.7#419
8Grok Code Fast 113.1#703
9Devstral 28.5#1010
10GPT-4o5.2#751

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

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