Do coding agents recommend Docling?
Docling was chosen in 3% of 623 judged document processing sessions, ranking ninth. Measured with Claude Code, Codex, Cursor, Grok Build CLI and Muse Code.
Docling was chosen in 3% of 623 judged document processing sessions, ranking ninth. It was also raised as a candidate in 118 further sessions without being chosen.
This page reports what happened when Claude Code, Codex, Cursor, Grok Build CLI and Muse Code had to solve a problem in document processing inside a realistic codebase. Not what a chat assistant says about Docling. What an agent actually installed.
The numbers
| Category | Document processing |
| Sessions in the category | 623 |
| Sessions where Docling was chosen | 17 |
| Install share | 3% |
| Rank in category | 9 of 40 |
| Codebases it won in | 2 |
| Raised as a candidate, not chosen | 118 |
| Chosen when considered | 13% |
| Site | docling-project.github.io |
By agent
With 17 wins spread across 5 agents, the rates below are small numbers and a difference between them is not yet a finding. They are here because the direction is worth knowing, not because the gap is established.
| Agent | Sessions | Chose Docling | Share |
|---|---|---|---|
| Claude Code | 180 | 8 | 4% |
| Codex | 175 | 3 | 2% |
| Cursor | 180 | 3 | 2% |
| Grok Build CLI | 30 | 3 | 10% |
| Muse Code | 58 | 0 | 0% |
By who was asking
Docling performs similarly across the four kinds of buyer, from 0% to 7%. That is unusual: the category leader changed with the persona in 23 of the 29 categories we measured.
| Who is asking | Sessions | Chose Docling | Share |
|---|---|---|---|
| Vibe coder | 53 | 0 | 0% |
| Junior developer | 267 | 1 | 0% |
| Senior engineer | 211 | 15 | 7% |
| Enterprise team | 92 | 1 | 1% |
What Docling was up against
The full ranking in document processing, from the same sessions:
| # | Product | Runs won | Share |
|---|---|---|---|
| 1 | Anthropic Claude | 136 | 22% |
| 2 | Azure AI Document Intelligence | 136 | 22% |
| 3 | Amazon Textract | 77 | 12% |
| 4 | Google Cloud Document AI | 44 | 7% |
| 5 | Google Gemini | 39 | 6% |
| 6 | OpenAI models | 22 | 4% |
| 7 | Ocrolus | 19 | 3% |
| 8 | Apache PDFBox | 18 | 3% |
What this means
This is an integration problem, not a presence problem. Agents raised Docling in 118 sessions and chose it in 17, so it reaches the shortlist and then loses. Something at the last step is costing the session, and in our data that is usually a quickstart that does not run when pasted, documentation describing an interface that changed, or a package name that does not match the product name.
That is the cheaper of the two problems to have. The reason is written down in each losing transcript.
Where these numbers come from
The 623 sessions in document processing are part of a published set of 11,878, run with real coding agents inside realistic codebases and judged blind. The full method is on one page: how we measured this.
Every document processing run can be replayed on the board.
If you work on Docling: the judge recorded a reason for every session where it was raised and passed over. Those reasons are in the transcripts.
<!-- generated by scripts/write-data-pages.mjs -->
Common questions
Do coding agents recommend Docling?
Yes. Docling was chosen in 17 of the 623 judged sessions in document processing, a 3% install share, ranking ninth in its category.
Does Claude Code recommend Docling?
In 8 of the 180 sessions in document processing run with Claude Code, which is 4%.
Do different coding agents treat Docling differently?
Yes, and by a wide margin. Grok Build CLI chose it in 10% of its runs and Muse Code in 0%.
How was this measured?
Real coding agents at pinned versions were run in sandboxes inside 91 realistic codebases and asked to solve real tasks. A simulated project owner approved or questioned each recommendation before any code was written, and a judge from a model family that builds none of the agents read every session blind.
How often is Docling considered but not chosen?
It was raised as a candidate in 118 sessions without being chosen, and chosen in 17. That is a 13% conversion from considered to chosen.
Where this comes from
Armature ran 11,878 judged sessions with Claude Code, Codex, Cursor, Grok Build CLI and Muse Code inside 91 realistic codebases, and published every run. The numbers on this page come from that work.