Last updated:
What is Jev AI? The non-text model from TypeSafe
Jev is an AI model that does not write. Built by TypeSafe AI and released on September 15, 2026, it answers typed questions with probabilities, a pick from a list or a score, which TypeSafe calls calibrated decisions. On October 9, 2026 the company raised $870 million at a $7.5 billion valuation.
What Jev is
- A new kind of model built on a transformer architecture, but not a large language model. It produces probabilities rather than generated text.
- TypeSafe positions it for automating tasks, not for writing text or code. Co-founder Diogo Almeida’s pitch: language models are great at human language, but computers "speak a different language".
- TypeSafe claims Jev runs significantly faster and uses far fewer tokens than LLMs. That is the company’s claim; we have not seen independent benchmarks.
- Requests go to an evaluate-style endpoint, not to a chat-completions API, so it does not drop into tools built around chat models.
Source: TechCrunch: maker of non-text AI model Jev valued at $7.5B (Oct 9, 2026)
Source: LiteLLM: TypeSafe Jev integration
Who makes it: TypeSafe AI and the $7.5B round
- Founded. 2024, by Diogo Almeida (previously an OpenAI researcher), Sasha Sheng (former Meta research engineer) and Erik Gafni.
- Launch. Jev was released on September 15, 2026 and spread quickly.
- Funding. $870 million at a $7.5 billion valuation, announced October 9, 2026. Andreessen Horowitz led; Sequoia and existing investor DCVC took part.
- Adoption. TypeSafe says a third of Fortune 500 companies already use Jev. TechCrunch reports this as the company’s claim.
Source: TechCrunch: maker of non-text AI model Jev valued at $7.5B (Oct 9, 2026)
How the Jev API works
You send a state and a set of typed questions to POST https://api.typesafe.ai/v1/systemone with a Bearer API key. The state can be plain text or structured data such as chat logs, records or application state. Answers come back keyed by your question IDs, with token usage.
| Question type | What you define | What Jev returns |
|---|---|---|
| Yes / no ("Noul") | A yes-or-no question | A probability from 0 to 1 that the answer is yes |
| Choice | Up to 255 options | The top choice, the full probability distribution and a confidence value |
| Score | An ordered rubric of 2 to 10 levels | A probability-weighted score that can fall between levels, plus confidence |
The model name is jev-latest (the current version is logged as jev-1.13). Check TypeSafe’s API docs for the exact request schema before you build on it.
Source: TypeSafe API docs
Source: LiteLLM: TypeSafe Jev integration
Jev pricing
- $0.042 per 1M input tokens, and $0 for output, as listed by LiteLLM and OpenRouter (Jev 1.13).
- TypeSafe’s own API page does not show a price or a free tier, so confirm on your TypeSafe account before you budget.
- For scale: GPT-6 Luna, OpenAI’s cheapest GPT-6 model, lists at $0.10 / $0.50 per 1M input / output tokens.
Source: LiteLLM: TypeSafe Jev integration
Source: OpenRouter: Jev 1.13
Jev vs an LLM
Where it fits, judging by the question types: sending a support ticket to the right queue (choice), flagging a message for review (yes / no) or rating a lead or an application against a rubric (score). Anything where you need words back is still LLM work.
| Jev | Typical LLM | |
|---|---|---|
| Output | Probabilities, a choice or a score | Generated text (or code) |
| Good for | Routing, classification, yes/no checks, scoring | Writing, chat, coding, summarising |
| Price model | $0.042 per 1M input, output free | Pay for input and output tokens |
| API | Own evaluate endpoint | Chat completions |
AI model API pricing · GPT-6 Sol and Luna
FAQ
What is Jev AI?
Jev is a transformer-based AI model from TypeSafe AI that answers questions with calibrated probabilities, a choice or a score instead of text. It is built for automating decisions and was released on September 15, 2026.
Is Jev an LLM?
No. It uses a transformer architecture but does not generate text. It returns a yes/no probability, a pick from up to 255 options with the full distribution, or a score on a rubric.
Who makes Jev?
TypeSafe AI, founded in 2024 by Diogo Almeida, Sasha Sheng and Erik Gafni. On October 9, 2026 it raised $870 million at a $7.5 billion valuation in a round led by Andreessen Horowitz.
How much does Jev cost?
$0.042 per 1M input tokens with no charge for output, according to LiteLLM and OpenRouter. TypeSafe’s API docs do not list a price, so check your account.
Can Jev write text or code?
No. TypeSafe positions it for automating tasks, not for writing text or code. Use an LLM for anything that needs words back.
Independent guide, not affiliated with TypeSafe AI. Facts come from TechCrunch’s October 9, 2026 report, TypeSafe’s API docs, LiteLLM and OpenRouter, checked on October 10, 2026. Adoption and speed figures are the company’s own claims.