AI Applications Leaderboard

AI applications tested on real-world legal work.

Results updated August 2026

Substance: Whether the legal work is correct, complete, and responsive. Measured as task pass rate, the percentage of tasks for which every applicable criterion was satisfied.
Form (1 to 3): Whether a lawyer would want to use the delivered work as written, including clarity, structure, and right-sizing.

Results, contract workflows

Contract Workflows results: run pass rate and form for each application.

Claude Cowork (Fable 5)	Anthropic	33.9%	2.61
Claude Web App (Fable 5)	Anthropic	32.3%	2.53
Claude Web App (Sonnet 5)	Anthropic	29.0%	2.57
ChatGPT Web App (GPT 5.6 Sol)	OpenAI	27.4%	2.75
ChatGPT Codex	OpenAI	25.8%	2.77
Gemini Web App (Gemini 3.1 Pro)	Google	14.5%	2.33
Microsoft Copilot Web App	Microsoft	12.9%	2.39

Key takeaways

Claude led on substance. Claude Cowork recorded the highest contract-workflow pass rate, while the Claude Web App with Fable 5 led data extraction and the combined benchmark.
ChatGPT Codex led on form in both categories. It produced the clearest, best-structured and most appropriately sized deliverables.
Gemini 3.1 Pro's substantive pass rate trailed every tested Claude and OpenAI application. That pattern held in both contract workflows and data extraction.
Microsoft Copilot recorded the lowest substantive pass rate in both categories. It also had the lowest form score overall across the two categories.

How we test

We gave each application the same legal tasks twice through its web app, the same interface its users work in, and preserved its response and native deliverables. Task-specific binary criteria determine the reliability pass rate, and a run that returns no usable output counts as a failure. A separate two-judge panel scores clarity, length, and structure from 1 to 3. We withhold flagged runs until a human validates them.

Read the full methodology