A San Francisco-based startup built an artificial intelligence model that refuses to talk to people, and investors are lining up anyway. TypeSafe AI released its Jev model just over a week before publication alongside its seed round, and the launch video alone racked up roughly 40 million views on X. Within days, the company was fielding funding offers that would value it at more than $10 billion, a staggering leap for a firm that raised its seed round only recently.
That seed round tells its own story about how fast things moved. As Bloomberg reported, TypeSafe AI had raised just $40 million in seed funding led by DCVC, a round that valued the company at $200 million, according to the Financial Times. Investment offers valuing the firm at more than $10 billion emerged within roughly 10 days of Jev's public debut, the paper reported, a jump most enterprise software companies never see across their entire lifespan, let alone in under two weeks.
A Model Built to Never Say a Word
Jev is unusual because it can't chat at all, and that's by design. Unlike autoregressive large language models that generate text token by token, Jev takes unstructured inputs and evaluates predefined options in parallel, returning schema-constrained typed choices with calibrated probability scores, according to Wikipedia's entry on the model. The system is meant to be read by other software, not by people, spitting out structured decisions rather than conversational replies.
The company's founder has deep roots in the technology he's now positioning against. TypeSafe AI was founded in 2024 by Chief Executive Diogo Almeida alongside Erik Gafni and Sasha Sheng, following Almeida's time at OpenAI, where he worked on reinforcement learning, InstructGPT, ChatGPT, and GPT-4, per Wikipedia. Almeida has questioned whether human-facing chat interfaces are necessary for back-end computer automation.
Where the Name Comes From
The branding leans on economic and psychological theory rather than typical tech-naming conventions. According to TrueFoundry, the model's name draws from 19th-century economist William Stanley Jevons, known for Jevons' paradox — the idea that increased efficiency leads to higher overall consumption — while its “System One” label nods to psychologist Daniel Kahneman's concept of fast, intuitive cognition. The framing captures TypeSafe AI's central bet: drastically lowering the cost of automated decisions will drastically expand how much of it gets used.
That pricing pitch is aggressive by any standard. TypeSafe AI prices Jev at $0.042 per million input tokens with no charge for output tokens, and the company claims the system runs 100 to 200 times faster and significantly cheaper than general-purpose LLMs on classification tasks, according to TrueFoundry and Linas's Newsletter. Output tokens are free because the model emits probabilistic choice indices instead of generating streams of text.
Fast Adoption Across Developer Platforms
Developers didn't wait long to plug it in. Within one week of its public release, Jev had been integrated into major developer distribution platforms including OpenRouter, Vercel's AI Gateway, and Cloudflare Workers AI, according to Linas's Newsletter. Developers can access Jev through those platforms.
Jev is designed for basic binary or multiple-choice decisions. Developers have historically had to force expensive, general-purpose LLMs to output JSON strings just to make basic binary or multiple-choice decisions, a workaround Jev is designed to replace outright.
Embarcadero Roots and In-Office Culture
TypeSafe AI is based in San Francisco and requires its engineering team to work in person five days a week, according to the company. That team reportedly includes former talent from Google Brain, Meta/FAIR, Stripe, and Airbnb, anchoring the young startup squarely within the city's dense AI hiring pipeline.
Jev's rapid rise raises a broader question about whether enterprises might use specialized, non-chat decision models for tasks now handled by general-purpose AI.
