Compare & Ask AI
Start from an example, or tick sections in the outline below. Compare shows them side by side. Ask AI sends only those sections, with their page numbers, so the answer can cite pages. Click a cited page to jump to it in the reader.
Why select sections first
The whole filing is about 317,000 tokens. Its outline, with page numbers, is under 4,000. A person or an AI agent can read the outline, pick the sections a question needs, and send only those.
Token counts are estimates (words × 1.3).
Read the Filing
Pick a heading to read that section in full, with its tables, images and page breaks. Filter the outline to find a topic. Tick sections to compare them or ask about them above.
Some figures are blank in the filing itself, for example “revenue growing from $0 at commercial launch in the first quarter of 2023 to $ for the three months ended September 30, 2026.” The S-1 as filed leaves those amounts blank, usually for the most recent quarter, and a later amendment fills them in. AILANG Parse keeps the text as filed.
Loading…
Every Table and Image
Every block is labelled with its type, so listing all the tables or all the images is a simple filter. Each entry shows its section and its page.
Some of the filing is only in pictures. The founders' letter that opens the prospectus is eight page images, and the customer case studies are images too. Each one links to the original on EDGAR.
Every Risk Factor, With Its Page
Each risk factor's title is a heading under its category, so one query returns the full list with pages:
The Result
This is the S-1 as filed on EDGAR, parsed by AILANG Parse with no AI model involved. Everything on this page comes from that parse: the sections, tables, images and page numbers.
What You Get
Every heading, paragraph, list, table and image comes out as a block labelled with its type, in reading order. A marker before each page gives its printed page number.
anthropicpbc-sx1.htm.json page 27
{"type": "section",
"kind": "page-break", "name": "27"}
{"type": "heading", "level": 1,
"text": "Risk Factors"}
{"type": "text",
"text": "Investing in our Class A common
stock involves a high degree of risk.
You should carefully consider …"}
{"type": "heading", "level": 2,
"text": "Risks Related to our Business
and Industry"}
{"type": "heading", "level": 4,
"text": "Our success depends in large
part on our ability to continue to
develop the most advanced and
capable AI models, …"}
{"type": "list", "ordered": false,
"items": ["advance the frontier of AI
capabilities through breakthrough
research, …", …]}
Run It Yourself
- Download the filing. SEC asks scripts to say who they are, so put your name and email in the
-A(User-Agent) value.curl -sL -A "Your Name you@example.com" -o anthropicpbc-sx1.htm \ https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm - Parse it. This writes
anthropicpbc-sx1.htm.json(the structure) andanthropicpbc-sx1.htm.md(Markdown to give a model).docparse anthropicpbc-sx1.htm - Query the JSON. For example, the outline, every risk factor with its page, or one row of a table:
jq -r '.document.blocks[] | select(.type == "heading" and .level <= 2) | (" " * (.level - 1)) + .text' anthropicpbc-sx1.htm.jsonjq -r 'foreach .document.blocks[] as $b ({page: "", h1: ""}; if $b.type == "section" and $b.kind == "page-break" then .page = ($b.name // "") elif $b.type == "heading" and $b.level == 1 then .h1 = $b.text else . end; select(.h1 == "Risk Factors" and $b.type == "heading" and $b.level == 4) | "p.\(.page) \($b.text)")' anthropicpbc-sx1.htm.jsonjq -c '.document.blocks as $b | ($b | map(.text) | index("Consolidated Statements of Operations Data")) as $i | $b[$i:] | map(select(.type == "table")) | .[0].rows[1]' anthropicpbc-sx1.htm.json # ["Revenue","$41","$386","$4,585","$1,163","$16,286"]
The same works for any EDGAR filing: 10-K, 10-Q, S-1 or proxy statement. See HTML parsing for more, and chunking to split the result for retrieval.