Tesla quarterly vehicle production and deliveries by model group

Vehicles produced and delivered each quarter by reported model group, Q2 2019 to Q2 2026 — 29 quarters, 174 rows — read from Exhibit 99.1 of Tesla's quarterly Production & Deliveries Form 8-K filed with the SEC.

The walkthrough of how the source was found and built is site/docs/examples/tesla-source-discovery.md.

Rebuild

SEC_USER_AGENT="<project> <your-email>" node fetch.ts   # network; reuses archive/ unless --refresh
node build.ts                                           # offline; archive/ → data/
node --test                                             # 32 tests
node chart.ts                                           # offline; data/ → the chart in the worked example

fetch.ts is the only script that touches the network, and it refuses to fetch without SEC_USER_AGENT (with a snapshot already in archive/ it reuses it and exits without needing one) — the SEC requires requests to identify the requester and asks for no more than 10 per second. build.ts reads archive/ only, checks every file against the SHA-256 in archive/manifest.json, and is deterministic: two runs produce byte-identical CSVs.

What the numbers mean

Deliveries are vehicles transferred to a customer with the paperwork complete. Production is vehicles built. They are two separate measures, they are never equal, and neither is revenue. Releases through Q4 2021 called the delivery count "slightly conservative" and said final numbers "could vary by up to 0.5% or more"; releases from Q1 2022 onward drop that language entirely.

Over this range Tesla does not report individual models. It reports groups, and the grouping has changed twice. (The earlier prose releases did give per-model figures — see below — so this is a property of the table era, not of Tesla's reporting for all time.)

Group labels are kept exactly as the release writes them. A combined group is never split into individual models, and groups are never re-combined across regimes, because the source does not support either. If you want a continuous series, Total is the only group that runs unchanged through all 29 quarters.

For a continuous series, use data/tesla-quarterly-totals.csv. It holds one row per quarter with the release's Total deliveries and production side by side, copied from the Total rows of the main resource by build.ts (a test checks they match). It is what the chart on this page plots.

is_total matters. Each quarter and metric has its component rows plus the release's own Total. Filter on is_total — summing a Total with its components double-counts. build.ts asserts that the components sum to the reported Total for every quarter and metric where the release printed a figure for every group, so if that ever stops being true the build fails rather than publishing the discrepancy. Where a group is dashed instead, the exact sum is unavailable by construction; the build still requires the known components not to exceed the Total, and the dash-resolution rule below is what decides the rest.

Empty is not zero. An empty vehicles cell means the release gave no figure. A 0 means it reported none. The one real zero is Model S/X production in Q1 2021, which Tesla printed as a dash during the line changeover; it is resolved to 0 only because the quarter's total minus the other group leaves exactly nothing, and a dash the arithmetic cannot account for is left empty instead.

Coverage, and what is not here

Coverage starts at Q2 2019 because that is the first release to publish its figures in a table. 19 earlier releases were found, covering 18 quarters between Q1 2013 and Q1 2019 — the run is not contiguous, and no delivery 8-K was filed at all for Q2 2013 through Q4 2014. They state the numbers in prose — "Q1 deliveries totaled 29,980 vehicles, of which 11,730 were Model S, 10,070 were Model X, and 8,180 were Model 3" — with the wording changing release to release. Those filings are archived and listed in data/source-filings.csv with layout: prose and extracted: false. They are a known gap, deliberately left rather than extracted by a set of one-off regexes whose failures would be silent. Note what that gap costs: the prose releases break out individual models (Model S, Model X, Model 3 separately), so extracting them would add finer granularity, not just more quarters. source-filings.csv accounts for all 50 archived exhibits — 29 table, 19 prose, 2 exhibits that are not delivery releases at all — so what was found and not used is visible rather than absent.

Other limits:

• Q3 2026 is not here. The snapshot was taken on 2026-09-18, before the quarter ended. The cutoff is Q2 2026, the latest quarter with a released report.
• The lease-accounting column is not published. Most releases carry a "Subject to operating lease accounting" percentage. It is a share, not a vehicle count, so it is out of this schema's grain; it remains in the archived exhibits.
• Energy storage deployments are not published, for the same reason — GWh, not vehicles.
• Three periods are covered by more than one exhibit: Q2 2017 was filed twice, an unrelated exhibit was filed the day before the Q3 2018 release, and the Q4 2022 filing carried a second exhibit announcing Investor Day. The build applies a preferred-version rule — the later filing wins, both snapshots stay in archive/ — but no quarter actually triggers it, because in each case only one exhibit carries a table.

Restatements: the annual recap does not match the four quarters

data/tesla-annual-deliveries.csv holds the full-year figures that Tesla recaps in each Q4 release. It covers 2020–2025: the Q4 2019 release carried no full-year recap, so the annual series starts a year after the quarterly one. They are kept in a separate resource because they do not reconcile with the four quarters as originally reported:

Tesla revises quarterly figures after the fact and does not reissue the press releases, so the annual number counts vehicles the original quarterly releases did not. build.ts reports these differences on every run rather than asserting them away or quietly preferring one figure. Both are what the source said at the time; which one you want depends on whether you are studying what was reported or what was finally counted.

Licence

The compilation here — the extracted figures, the schema and the build scripts — is released under PDDL-1.0.

The sources are Tesla press releases filed with the SEC as public records. The figures themselves are facts, which US copyright does not cover; the press-release prose is Tesla's, and this dataset does not republish it. No redistribution licence is granted or implied for the source documents, and none was found: neither Tesla nor the SEC publishes terms that grant one. archive/ holds the retrieved documents as evidence for the build, not as a republication of them.

Tesla's investor-relations site and the Business Wire copies of the same releases return HTTP 403 to plain clients. SEC EDGAR serves the identical exhibits as public records and states its access terms openly, which is why it is the source used here.