Agent leaderboards / All sectors / Document processing & OCR
Document processing & OCR: which document products coding agents choose
Azure AI Document Intelligence won about 24% of 801 runs.
Read this leaderboard as textrankings, key learnings, method
Key learnings
We asked five agents to read the documents that arrive at 12 codebases, 801 runs in all, in several wordings and as four kinds of user. Anthropic Claude came second with about 19%, and Amazon Textract third with about 13%.
Each agent had its own favourite
Claude Code picked Anthropic Claude in 112 of its 180 runs. Codex, Cursor and Muse Code all led with Azure AI Document Intelligence. Grok Build CLI led with Google Gemini, seven of its 30 runs.
The same question in other words got other answers
A case is one codebase with one agent, asked again in different words and as different people. 53 of 60 cases did not land on one product every time.
- The simulated user, who approves each plan, sent the agent back at least once in 645 of 801 runs.
- Enterprise teams split almost evenly between Amazon Textract at 41 wins and Azure AI Document Intelligence at 38.
- Codex picked its own maker's OpenAI models in seven of 235 runs, and Cursor never picked xAI Grok.
- Tesseract OCR was named 391 times and never chosen, which is no loss, since it is an engine rather than a service.
The ranking801 runs
By agent, by persona, by wording
By agent
| Codex235 runs | Azure AI Document Intelligence · 75then Google Cloud Document AI · 42 |
| Claude Code180 runs | Anthropic Claude · 112then Azure AI Document Intelligence · 20 |
| Cursor180 runs | Azure AI Document Intelligence · 47then Google Gemini · 27 |
| Muse Code · Muse Spark 1.3176 runs | Azure AI Document Intelligence · 46then OpenAI models · 34 |
| Grok Build CLI · Grok 4.730 runs | Google Gemini · 7then Azure AI Document Intelligence · 5 |
By persona
| Junior developer327 runs | Anthropic Claude · 84then Azure AI Document Intelligence · 62 |
| Senior engineer270 runs | Azure AI Document Intelligence · 79then Anthropic Claude · 45 |
| Enterprise team136 runs | Amazon Textract · 41then Azure AI Document Intelligence · 38 |
| Vibe coder68 runs | Anthropic Claude · 16then Google Gemini · 16 |
By what the ask stressed
| The plain ask788 runs | Azure AI Document Intelligence · 193then Anthropic Claude · 150 |
A case is one codebase with one agent, asked several times in different words and as different people. 53 of 60 cases did not hold to a single document product.
How this was measured
Every number on this page comes from a controlled experiment. We took 12 small applications, asked 5 coding agents (Codex, Claude Code, Cursor, Muse Code (Muse Spark 1.3), Grok Build CLI (Grok 4.7)) to read the documents that arrive at each of them, in several wordings and as a junior developer and senior engineer and enterprise team and vibe coder, and let the agent choose the product. Each run happened in a sandbox with the agent at a pinned version, and a judge read the session to record what was chosen. That is 801 runs. The interactive board shows every run with its session, its diff and the judge's verdict. A simulated user stood in for the owner of the codebase: it read the agent's plan and had to approve it before any code was written; it sent the agent back at least once in 645 runs. Read the methodology and the publications.
If you sell in this sector: what these numbers mean for a vendor.
Open the interactive boardThis page as MarkdownOther sectors
- Agent sandboxes
- Observability
- AI SRE
- Payments
- Deploy
- Auth
- Email providers
- Product analytics
- Databases
- File storage
- LLM evals & observability
- Voice Agents
- Serverless functions
- Cloud
- AI gateway
- Bot protection
- Search
- Agent frameworks
- Performance in CI
- Usage-based billing
- Internationalization
- Security testing
- Maps
- Message queues
- AI search
- Code review
- E-signature
- Security platforms
- In-app chat & calls
- Vector search