# TypeSafe AI exits stealth with $40M and Jev

> TypeSafe AI has released Jev, a model built for structured software outputs, alongside a $40 million seed round led by DCVC.

- Publication: Go Big News
- By Nifal Adam, Go Big News Staff (Staff News)
- Category: AI
- Published: 2026-09-23T08:49:31.102+00:00
- Canonical: https://www.gobignews.com/ai/typesafe-ai-exits-stealth-with-40m-and-jev

---

[TypeSafe AI](https://typesafe.ai) emerged from stealth on September 15 with a $40 million seed round led by DCVC and the release of Jev, its new AI model. CEO Diogo Almeida said he spent two years building the model and its training method.

Jev uses RLCD, or Reinforcement Learning for Calibrated Decisions. TypeSafe describes RLCD as optimizing for calibrated decisions, with probabilities it characterizes as epistemically honest, on what it calls System One tasks, in contrast with RLHF, the method it says trains existing chat LLMs.

In a [post on X](https://x.com/completeskeptic/status/2099925682726002904), Almeida said Jev is 20 to 200 times faster than chat LLMs; the post's second figure, 40 to 400 times, is cut off mid-line.

TypeSafe's [launch blog post](https://typesafe.ai/blog/introducing-system-one-models-and-jev) puts Jev's end-to-end response time at 70 milliseconds to 500 milliseconds, compared with 3 to 329 seconds for its selected frontier LLM benchmarks: GPT-5.6 Terra, GPT-6 Astra and Fable 5.1. The company says that comparison translates to a 40 to 200 times speed advantage for System One-shaped queries. It did not cite an independent benchmark.

The company lists Jev input pricing at $0.042 per million tokens, or $42 per billion tokens, with output tokens free. Its stated LLM input-price comparison ranges from $0.20 to $10 per million tokens.

Rather than producing chat text token by token, Jev returns typed, structured values with probability and confidence information for software use. The Register reported that TypeSafe showed a customer-service routing demo in which Jev responded in 0.114 seconds, versus 8.566 seconds for GPT-5.6 Terra on the same task.

Jev is available through an early-access waitlist. TypeSafe identifies Almeida as a co-inventor of RLHF and InstructGPT; its leadership team also includes COO Sasha Sheng and CTO Erik Gafni. DCVC general partner James Hardiman said the company is addressing the challenge of making capable models dependable enough for developers to build into products at scale.

---

## Companies in this story

- Typesafe: https://www.gobignews.com/company/typesafeai

---

## X profiles in this story

- @completeskeptic: https://www.gobignews.com/profile/completeskeptic