Docling is an open-source document conversion toolkit started by the AI for knowledge team at IBM Research Zurich and hosted as a project of the LF AI & Data Foundation. The code is MIT-licensed, the latest release on PyPI is 2.139.0, and the project has about 69,000 GitHub stars (Docling on GitHub, PyPI, accessed 15 October 2026).
It is good at what it does, and it costs nothing to license. If you are looking for a Docling alternative, anyformat solves an adjacent problem: schema-validated extraction wrapped in workflow orchestration, human review and a managed service.
anyformat is enterprise-ready agentic document processing: agentic OCR that turns any document into data you can trust, where every value is validated and cites its source. It is built for companies with serious demands on security, scale and audit, so process owners and their IT teams use it alongside the developers who evaluate and integrate it.
You describe the fields once as a schema. Parse reads the document, Extract returns each value with a confidence score and its position on the page, Validate runs your rules, and low-confidence values go to a reviewer in the same workflow, through the API, the Python and TypeScript SDKs or the no-code Studio. Every value carries its evidence, it holds up on hard documents, it is a European company with ISO 27001 and GDPR, and the price covers the whole workflow, not only the extraction. Compared with a library you run yourself, those steps come in a managed service.
The measure that matters is how many documents go through without anyone touching them, and teams go live in days, not quarters, with SDKs, templates and evals to measure accuracy as they go.
What Docling is, and what it is not
Docling reads many formats into one unified representation and exports it. Inputs include PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, email, and audio and video with a transcript; outputs are Markdown, HTML, lossless JSON, plain text and chunks for RAG (Docling docs, accessed 15 October 2026). For PDFs it handles layout, reading order, tables, code, formulas and OCR for scans.
It plugs into LangChain, LlamaIndex, CrewAI and Haystack, ships a command-line tool and an MCP server, and can run as an API through docling-serve. It is a toolkit you assemble and operate: there is no vendor selling it as a service, and everything around the conversion step is yours to build.
Customization and extraction approach
Docling's main job is conversion: the output is a document (Markdown or JSON) that a later step turns into the fields you need. The docs also list a structured data extraction feature, marked as beta (Docling docs, accessed 15 October 2026). Secondary sources describe it as experimental, limited to PDFs and images, and driven by a template you define; check the official docs for its current state before depending on it.
anyformat extracts against a user-defined schema in zero-shot and returns validated JSON, with document-level and table-level fields declared together and returned in one call. Every field carries a confidence score and a visual citation pointing at its exact position on the page.
Confidence: how well it converted versus whether the value is right
Docling reports confidence at page level and at document level, built from layout, OCR and parse scores between 0 and 1, grouped into four grades from poor to excellent. The table score is listed as not yet implemented, and the docs say the numerical scores are informational and their computation may change (Docling docs, accessed 15 October 2026).
That tells you how well a page was converted. It does not tell you whether an extracted value is right, so a review process has nothing to route on at the field level. anyformat returns a confidence per extracted field, with the source evidence next to it, so a threshold decides which values flow through and which go to a reviewer in the same workflow. For how the scores are produced and read, see how confidence scoring works.
Benchmark: where each one leads
In our own parsing benchmark, run on 1,000+ real documents across 30+ document types and scoring text, tables and document structure, anyformat scored 78.1% and Docling scored 52.3% (our benchmark, 2026). The same table lists Docling as self-hosted, so it has no per-page price. anyformat also publishes its own parsing benchmark with the full leaderboard.
On the independent ParseBench leaderboard, read on 15 October 2026, anyformat (Standard) scored 80.30, fifth of 147 entries, and the open-weight Docling models scored 50.65, 111th. ParseBench is published by LlamaIndex, which also sells a parsing product that tops its own leaderboard, and the Docling row is the open-weight models, not necessarily a tuned full pipeline.
Docling holds up better in a hands-on comparison. In our test of PDF-to-markdown tools on the same fictional files, Docling returned the table as a correct markdown table, the list as a list and the two columns in order, and read a scan with no configuration. We compared by eye and saw no errors; that is an observation, not an accuracy score (PDF to markdown for LLMs, 30 September 2026). Benchmarks also measure parsing, not extraction against your own schema, which is why a test on your own documents is the real comparison.
Accuracy: reading, structure and values
Accuracy means three different things in document AI. Reading the characters and rebuilding structure (tables, reading order, formulas) is Docling's core, and its Markdown output is strong there. Getting the right value into the right field is extraction, and it is what a downstream system consumes: a conversion score cannot tell you whether the invoice total landed in the total field, or whether row 1,800 of a long table survived. anyformat's long-document study held about 99% row recovery on tables running to 50 pages and roughly 2,400 rows (our benchmark, 2026).
If accuracy decides the choice, test it the way you will run it. Take 50 to 100 of your own documents, extract the same schema with both, and score at field level: exact match for identifiers and amounts, row recall for tables. Then check whether the confidence each system reports predicts which fields are wrong, because that decides how much a person has to review. anyformat's evals suite runs that kind of comparison on your own documents before a workflow goes live.
Run both on your own documents. Try anyformat free, with a one-time grant of 50,000 credits on Pay As You Go, or book a demo.
Docling has no license fee: the code is MIT and Granite-Docling, its small vision model, is Apache 2.0 (Granite-Docling model card, accessed 15 October 2026). The cost is yours to run it: infrastructure, engineering time and everything built around conversion. The docling-serve container images are large, with a base image of 8.7 GB on amd64, and models download on first use (docling-serve, accessed 15 October 2026).
anyformat bills in credits, charged per page and per operator: 10 credits cost €0.01, so Parse is 25 credits per page (€0.025, published as $25 per 1,000 pages), Extract 35 credits per page, and Validate 5 credits per AI rule, once per extraction (anyformat pricing, 2026). Parse plus Extract is 60 credits, or €0.06, per page.
On licence cost alone Docling is cheaper, since it is free. The two are not the same product: anyformat's price includes the managed service, the review queue, the validation and the orchestration that you would otherwise build and run yourself, and its Free tier lets you test before paying.
Compliance and European sovereignty
Docling is software, not a vendor, so it carries no certifications of its own: compliance depends on where and how you run it. That can be an advantage when you need full control of the environment, and a workload when an auditor asks for evidence.
anyformat is ISO 27001 certified, GDPR-compliant and an EU-headquartered company with an EU legal entity, with compliance included at every plan and no surcharge.
Workflow builder, review and validation
Docling converts documents; classification, splitting multi-document packets, routing by document type, validation rules and human review are not part of it, and its documentation describes no review queue. Teams that choose Docling build those steps around it.
anyformat includes Studio, a no-code workflow canvas: branching, conditions, splitting, routing, extraction operators, validation rules, cross-referencing against internal data sources, and human-in-the-loop review placed wherever the process needs it. Low-confidence fields route to a reviewer, who sees the value next to its source on the page and can correct it. Operations teams change the flow without a code deploy.
Version stability
With a library you pin the package version yourself, and the models it downloads are part of what you have to keep consistent between environments and over time.
anyformat workflow versions are pinned and immutable: a published version_id does not change underneath you, which matters when a regulated process has to reprocess a document identically years later.
Yes, if you want extraction to run as a process rather than a call: schema-validated output with per-field confidence, a review queue, validation rules, no-code orchestration and version-pinned workflows, without running models and containers yourself. If you want markdown or text out of documents, at no license cost and on your own machines, it depends on the job: Docling does that well, and anyformat is not a cheaper way to do it. And if you like Docling's conversion, the two can sit together, because the layer that makes the output production-safe is the part anyformat adds.
When to choose Docling
Free, open-source conversion of many formats into Markdown or JSON, with the pipeline around it as code your team is happy to own, or when your constraints require running everything on your own machines.
When extraction has to run as a process rather than a call: per-field confidence with evidence, validation rules, human review, no-code orchestration and pinned workflow versions, as a managed service. Get started at anyformat.ai.
See also the other comparisons: anyformat vs LlamaParse, anyformat vs Unstructured and anyformat vs Mistral OCR, or browse all comparisons.
