Skip to main content
AI Market Watch
Loading...

TypeSafe AI

Category: Foundation Models / LLMs

TypeSafe AI is a San Francisco frontier AI lab building machine-native, composable AI models that return typed, calibrated decisions instead of text. TypeSafe AI was founded in 2024. The company is led by Diogo Almeida. Based in San Francisco, United States. Team size: 11-50. Total funding raised: $40M. Latest round: Seed. Key investors include DCVC, AWS Startups.

Founded
2024
Headquarters
San Francisco, United States
Team size
11-50
Total funding
$40M

Value proposition

Provides developers with fast, cheap, reliable machine-native intelligence that can be embedded as a software primitive for semantic judgment and automation — no hallucinations, calibrated confidence scores, and typed outputs software can act on.

Products and solutions

Jev — the first "System One Model," a transformer-based model that outputs typed probabilistic decisions (choice, score, noul) with calibrated confidence instead of text, available via API, TypeSafe SDK, and hosted on Vercel AI Gateway. Trained via Reinforcement Learning for Calibrated Decisions (RLCD) on synthetic data.

Unique value

A non-chat, non-LLM decision model that cannot hallucinate text, delivers frontier-level intelligence at <100ms latency, and is up to ~100-445x faster and cheaper than frontier LLMs, with calibrated confidence for autonomous vs. human-review decisions.

Target customer

Software developers, engineering teams, and agentic-infrastructure/platform companies building automation workflows, model routing, and safety classifiers.

Industries served

AI infrastructure, software automation, developer tools, agentic AI

Technology advantage

New model class (System One Models) with a novel architecture, sampler, and training algorithm (RLCD — Reinforcement Learning for Calibrated Decisions); trained exclusively on proprietary synthetic data; typed outputs with calibrated probabilities; output tokens free and input tokens metered per billion.

How they differentiate

Unlike LLMs optimized for human language, Jev is built for machines — it returns typed decisions with calibrated confidence rather than text, eliminating hallucinations and enabling deterministic-style integration into software at dramatically lower cost and latency.

Main competitors

OpenAI (GPT/Luna models used for classification/routing), Google Gemini, open-source typed-decision alternatives built on open-weight LLMs

Key partnerships

Vercel (Jev hosted on Vercel AI Gateway), AWS Startups (investor/partner)

Notable customers

Vercel, Bryo AI

Major milestones

Founded 2024, two years in stealth, emerged from stealth Sept 15, 2026 with $40M seed led by DCVC, launched Jev, the first System One Model.

Growth metrics

Emerged from stealth Sept 15, 2026; demand so high at launch the API briefly lost ability to serve users; Vercel reported Jev used by ~13% of paid teams within 24 hours of launching on AI Gateway (2x GPT-5.6 family, 6x Fable 5.1); ~$200M valuation at seed.

Market positioning

Emerging category creator in machine-native / composable AI; positioned as a complement and partial replacement for LLMs in automation, routing, and safety-classification use cases.

Geographic focus

United States (San Francisco); global developer market

About Diogo Almeida

Ex-OpenAI Researcher (co-inventor of RLHF, InstructGPT, ChatGPT, GPT-4); Ex-Google Brain. Georgia Institute of Technology.

Latest news about TypeSafe AI

More Foundation Models / LLMs companies

Official website:

TypeSafe AI - AI Startup Profile | AI Market Watch