sec-api.io alternative

Their published coverage runs from 1993 to the present at 20 million filings. They list more than 40 APIs, plus a bulk dataset library they size at over 1,000 terabytes. New filings are indexed as EDGAR publishes them, and they quote under 300 milliseconds for N-PORT.

PIT holds 563,594 SEC rows across ten months, counting 18 form types. Certified means we hold the day and can say what we read on it. Five of those months run consecutively, 2022-11 through 2023-03, and every backfilled source stops on 2023-03-31. That is roughly one row for every thirty-five of theirs, over ten months against thirty-three years.

Their Query API documents filedAt as the Accepted attribute from EDGAR, in ISO 8601 with an Eastern Time offset. Their docs note that Accepted and Filing Date can land on different dates. That is a genuine acceptance timestamp at second resolution on every row.

PIT stores the same field as acceptance_at. It is populated on 3,124 of 563,594 SEC rows, or 0.55%, all of them on 2023-03-10. That is the one day we hold a header tape for. On the other 99.45% it is null, and published_at carries day resolution stamped at 23:59:59Z. For the question of when a filing was accepted, their row is better evidence than ours on every day but one.

A date range on filedAt is a real filter on acceptance time, and for plenty of research questions it is the correct one. It still reads the index as it stands today. A row that arrived late, whether through a delayed index entry, a re-crawl or a backfill, comes back inside a range that ended before it arrived. The clock on the row describes the filing, not the store. PIT also keeps committed_at, the instant a row entered our corpus. Cutting on it asks what we held at a given time, rather than what we now say was filed by then. The three selectable clocks are on filing timestamps.

An empty range has the same problem from the other side. Their filing object, as the Query API documents it, carries filedAt, periodOfReport, effectivenessDate and effectivenessTime, and the documentation describes no per-day completeness record alongside them. Ask PIT for a day we never ingested and the call returns HTTP 409 with count as JSON null and the day named in coverage.missing. Ask for a day we certified with nothing matching and it returns HTTP 200 with count: 0. Those are different HTTP statuses, so a client can branch on them. See coverage certificate.

A PIT row holds form type, company name, accession, CIK, five clocks, a content hash and the SEC URL. The filing text is not on it, so anything a language model writes about a cut comes from its own memory of the period rather than from the payload. Filing text is being ingested now; until it lands, fetch the document from source_locator yourself. sec-api.io sells the text itself, so if the work needs what a filing says, buy theirs.

• sec-api.io answers which filings exist matching a query; PIT answers what a cut held at an instant. They are different questions and PIT does not serve the first.
• Their filedAt is a real acceptance clock at second resolution on every row; our acceptance_at is null on 99.45% of SEC rows.
• The corpus holds no filing text, so nothing on this page compares full-text search.
• PIT’s certified backfill ends 2023-03-31 and the tape resumes 2026-08-26, while their index tracks EDGAR within seconds.
• Prices on this page are each vendor’s published list prices read in August 2026, not a live feed.
• Researcher $29 and Power $79 both read the whole ingested history. Plans and what the corpus covers.

No account needed. Backtesting an LLM on filings without lookahead works through the four leak mechanisms and the size of each.