TypeSafe opens Jev AI to public after rapid adoption forces waitlist removal
The startup's "System One" model processes decisions in milliseconds, but adversarial vulnerabilities raise questions about deploying it in automated agent systems.
Five days. That’s how long TypeSafe AI’s waitlist lasted before the startup decided to just open the doors to everyone.
On September 20, 2026, TypeSafe AI removed the access restrictions on Jev, its novel AI model designed not to write essays or generate code, but to make fast, structured decisions. The model first launched on September 15 with a waitlist. Developer demand apparently made that bottleneck untenable in under a week.
New users now get $5 in credits to start experimenting. At $0.042 per million input tokens, with no charge for output tokens, that $5 goes further than you might expect.
What Jev actually does, and why it’s different
TypeSafe AI calls Jev the first “System One” model, borrowing the term from Daniel Kahneman’s framework for fast, intuitive thinking versus slow, deliberate reasoning. Where most AI models aim to be general-purpose text generators, Jev is built for something narrower and arguably more useful in production environments: rapid probabilistic decision-making.
Instead of generating paragraphs of text, Jev accepts a program state alongside a typed query. It then returns structured outputs: user-defined choices, scores based on provided rubrics, or yes/no probabilities paired with calibrated confidence scores.
The model processes inputs at latencies between 70 and 500 milliseconds, with a context window of roughly 32,000 tokens.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
TypeSafe AI was founded by former OpenAI researcher Diogo Almeida, along with co-founders Gafni and Sheng. The company raised a $40 million seed funding round led by DCVC.
Rapid integrations signal market appetite
Since launch, integrations have gone live with LiteLLM, Netlify AI Gateway, Cloudflare, OpenRouter, and Vercel tools. That’s a significant spread across infrastructure providers in less than a week.
Independent assessments have confirmed strong performance in narrow classification tasks, validating TypeSafe’s pitch that purpose-built models can outperform general-purpose ones on specific workloads.
The adversarial problem no one can afford to ignore
Independent testing has revealed that Jev is vulnerable to adversarial text inputs that can steer its decision-making in unintended directions. Manipulated inputs or misleading cues can produce inaccurate outcomes or interfere with safe decision-making processes, potentially generating false negatives when the correct answer should have been a flag or alert.
Currently, no extensive production-level benchmarks or regulatory examinations have been documented following the model’s transition to open access.