Announced 9 Oct 2026 · Sources checked
What did TypeSafe publish on 9 October?
The company page is dated 9 October 2026 and titled “TypeSafe A raises Series AI.” The joke in the headline is the only place the post treats the raise as a Series A. The body says “a really big series A” led by Andreessen Horowitz, with Sequoia Capital, existing investor DCVC, “and the tech illuminati (aka a bunch of the best angel investors).” A numbered footnote is the only place the dollar amounts appear: “$870 million at a $7.5B valuation, with Martin Casado joining the board.”
The developer bullets promise more machine-native models and “everything else to be the best infrastructure for building smart software.” The business bullets say a third of the Fortune 500 “are getting their Jev on,” that TypeSafe has “saved customers millions of dollars in production already,” and that enterprise features are coming. Those savings and that customer count are TypeSafe’s. The post does not name a customer, a close date inside 9 October, or a second pricing change.
This is a priced round on a decision-model API, not a new checkpoint. For two other 2026 raises that also split company text from the wires, see Nous Research’s $90 million note and Butterfly Effect’s Manus round. For the hosted decision API that already prices itself against Jev, see Celeris-1-decision.
Which adoption numbers sit on which page?
a16z’s “Investing in TypeSafe AI” is dated 9 October 2026. It calls Jev “the fastest growing model we’ve ever seen – reaching 1 trillion tokens generated in merely 3 days post launch.” Later it says “25% of the fortune 500 enterprises have integrated Jev” and coins “the Jev-ons paradox.” The 25% line and the one-trillion-token line are Andreessen Horowitz’s. They are not on TypeSafe’s footnote.
TypeSafe’s own business list still says “a third.” TechCrunch, posted 2:41 p.m. PDT, repeats “a third of Fortune 500 companies” as a startup claim and dates Jev’s public launch to 15 September. SiliconANGLE, updated 16:13 EDT, uses “about a third of the Fortune 500” and “less than a month after TypeSafe launched Jev.” We are reporting both percentages rather than averaging them.
TechCrunch names co-founders Diogo Almeida, previously a researcher at OpenAI; Sasha Sheng, a former Meta research engineer; and Erik Gafni. It says the company was founded in 2024. Almeida’s 15 September launch post, which we opened the same evening, is signed “Diogo Almeida, founder, TypeSafe” and says he “helped build the methods that made language models useful at following instructions” at OpenAI. Those biographies are the writers’. We did not verify employment records.
What do the docs say Jev is, tonight?
The Models page we opened lists one current checkpoint: Jev 1.13, id `jev-1.13.0`. Price is $42 per billion input tokens, which the same table writes as $0.042 per million. Output tokens are free. Rate limits are 100,000 tokens per second and 80 requests per second, and the page says those limits “are adjusting dynamically.” Context is 64k tokens per request, with a 32k budget for `state` plus the longest question. Input is “Text only. String, JSON object, or array of text values. No image, audio, or video input.”
Aliases `jev-latest` and `jev-preview` both pointed at `jev-1.13.0` on the page we read. The API reference is `POST https://api.typesafe.ai/v1/systemone`. Questions are `noul` (yes/no probability), `choice` (at most 255 options), and `score` (two to ten rubric levels). Choice is capped at 255 options. The System One concept page says Jev “does not write replies, produce code, or generate explanations of its reasoning.” English is “the primary training language.” Customer requests “are not trained on,” the Models page says; enterprise zero-data-retention is pointed at a legal page we did not reopen as a contract.
That is a typed decision head, not a chat model. For the same noul / choice / score shape on another host, see OpenAI’s Decisions API. For an open-weight pair that also refuses to emit tokens, see Liquid AI’s open d1 models. For why a typed answer is a different object from generated text, see structured AI output.
| Field | What the docs print | What that is not |
|---|---|---|
| Price | $0.042 / M input tokens; output free | A negotiated enterprise rate |
| Context | 64k request; 32k for state + longest question | A 1M-token window |
| Input | Text, JSON, or text arrays | Images, audio, or video |
| Live id | `jev-1.13.0` behind `jev-latest` | A new 9 October checkpoint |
Which speed and cost lines are TypeSafe’s older launch claims?
The 15 September launch post is still the place TypeSafe prints the large multiples. It prices input at $0.042 per million tokens and output as free, matching tonight’s Models table. End-to-end response time is “70ms-500ms,” written as “40x-200x faster” than frontier models on “System One shaped queries.” A workflow-eval nuance block says the homepage’s “193.6x faster, 444.6x cheaper” pair comes from those workflows and “we expect that these are on the higher end of real world gains.” The same post says schema matching is “mathematically impossible” to fail and therefore plots a 0% type-error rate that “is not empirical.”
a16z restates the economics as “roughly 1/100 to 1/500 the cost of frontier models while being 100x faster for classification tasks at the comparable accuracy.” SiliconANGLE writes “under 700 milliseconds,” “up to 200 times faster,” and “up to 100 times more cost-efficient.” Those three pages do not share one harness. We timed nothing.
The launch post’s “can’t hallucinate” sentence is a claim about constrained outputs, not a claim that every probability is factually true. The System One docs say calibration “is measured across groups of predictions; it does not guarantee that an individual answer is correct.” Readers should keep those two TypeSafe pages in separate columns.
What is already on the decision-model shelf?
Cloudflare’s 9 October Clef-omni drop is an Apache-2.0 30B-A3B decision head that reads embedded audio and video on Workers AI and prices omni at $0.15 per million input tokens. Clef-flash’s hosted rate on that changelog is $0.038. Celeris’s 8 October hosted model lists $0.04 per million and a vendor jev-bench table. Microsoft’s same-evening Foundry model, Microsoft-Decision-1, prints $0.042 per million input tokens — the same list price as Jev 1.13 — on a post-trained Qwen3.5-9B. Those are three other vendors’ pages.
The 9 October raise does not add weights, a new Jev version, or a published customer appendix. TypeSafe says it will “take all of the things you love about Jev to the extreme” and add unspecified enterprise features. SiliconANGLE says proceeds go to more System One models. That is intention, not a shipping list.
A typed gate can still fail open. For a preprint that tells readers not to let typed decision models decide, see the option-channel paper.
What should a buyer check before treating $7.5 billion as a product fact?
Pin `jev-1.13.0` rather than `jev-latest` if a threshold was tuned on 1.13. The Models page says aliases move. Re-run the questions you already score — a support ticket, a policy clause, a log line — and compare probabilities with your current router. The 15 September workflow table is not a production SLA.
Ask which Fortune 500 figure the seller is using. TypeSafe’s “third” and a16z’s “25%” were both live on 9 October. Neither page names accounts. The one-trillion-token line is a16z’s, not a public telemetry dashboard we opened.
This is an evidence review of the 9 October company note, the a16z announcement, the TechCrunch and SiliconANGLE wires, the 15 September launch post, and the Models, System One, and API pages. We did not call `/v1/systemone`, buy a seat, or see a capitalization table.
Common questions
Did TypeSafe and a16z print the same Fortune 500 share?
No. TypeSafe writes “a third.” a16z writes 25%. SiliconANGLE follows TypeSafe. We did not reconcile them.
Did the 9 October raise change Jev’s price or checkpoint?
Not on the docs we opened. Jev 1.13.0 is still $0.042 per million input tokens, output free. The raise page does not announce a new model id.
Is Martin Casado’s board seat only on the wires?
No. It is in TypeSafe’s own footnote. a16z’s byline includes Casado. We did not see a Delaware filing.
What to remember
Use TypeSafe’s 9 October footnote for $870 million, $7.5 billion, and Casado’s seat. Use a16z for the lead and the 25% / 1-trillion-token lines. Use the Models page for `jev-1.13.0` and $0.042. Keep “a third” and “25%” in separate sentences.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. Cover photography is illustrative and does not depict the specific announcement or product.
Our editorial standards
