🔍 Read the full analysis: Jev And The Rise Of “System One” AI: Why The Next Useful Model Might Be One That Can’t Write A Sentence on ThorstenMeyerAI.com
Get the latest gadgets delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
TypeSafe launched Jev, a decision-focused AI model that produces typed, structured answers with probabilities, marking a shift from traditional text-generating language models. This approach aims to improve automation speed and reliability in enterprise settings.
On September 15, 2026, TypeSafe AI unveiled Jev, a new AI model designed for decision automation within software systems, eschewing traditional text generation. This development signals a significant shift in enterprise AI, focusing on structured, probabilistic decisions rather than language-based outputs, and aims to reduce costs and increase speed for internal decision-making processes.
Jev is part of TypeSafe’s broader concept of System One Models, inspired by Daniel Kahneman’s psychology, which prioritize quick, intuitive judgments over slow, deliberate reasoning. Unlike large language models (LLMs) that produce prose, Jev provides structured, typed answers such as ‘team: billing, confidence: 0.94,’ enabling software to act directly on decisions without parsing text. Built by Diogo Almeida, a co-inventor of RLHF at OpenAI, Jev employs a novel training approach called Reinforcement Learning for Calibrated Decisions (RLCD), designed to avoid issues like overconfidence and mode dropping common in LLMs.TypeSafe claims Jev can deliver responses in 70 to 500 milliseconds at a cost of roughly $0.042 per million tokens, making it significantly faster and cheaper than traditional models—up to 193.6 times faster and 444.6 times cheaper, according to the company. The model is marketed as producing ‘zero hallucinations’ in its outputs, meaning it conforms strictly to schema and cannot invent off-schema answers, though it does not guarantee correctness of decisions. Independent benchmarks show Jev’s accuracy varies depending on task complexity, with some tests indicating it performs well on narrow questions but less so on broader, more ambiguous ones.
Jev vs. LLMs: who should make the call?
Jev, from TypeSafe AI, is a “System One” model. It doesn’t write text. It returns a typed decision with a confidence score that your software can act on directly.
Same support ticket, two kinds of answer
“This ticket appears most likely related to billing, although it could also concern account settings or a recent plan change. I would suggest reviewing the invoice history before…”
A person reads it, or code has to parse the prose.
team: "billing"Software reads it and acts. Nothing to parse.
How they differ
| LLM | Jev | |
|---|---|---|
| Output | Text written for people | A choice, a score or a yes/no probability |
| Speed | Seconds per call | 70–500 ms* |
| Price | Input and (pricier) output tokens | $0.042 per million input tokens, output free* |
| Knows when it’s unsure | Often sounds confident when wrong | Confidence score on every answer |
| Explains its answer | Yes | No, which matters for audits |
| Best at | Reasoning, writing, open questions | Routing, tagging, scoring, duplicate checks |
* Vendor-reported. TypeSafe also claims up to 194× faster and 445× cheaper on its own selected workflows.
Accuracy is something you build
Jev is far cheaper and faster, but not more accurate than frontier models. How you phrase the question matters a lot.
TypeSafe’s benchmark scores agreement with two frontier models rather than verified ground truth. The five-question result used weights fitted on 1,000 labelled examples.
The real idea: a confidence dial you control
“duplicate listing”, confidence 0.62
Raise the threshold for fewer mistakes and more manual review. Lower it for more automation and more risk.
Only use Jev when all four hold
Good fits
- Routing tens of thousands of support tickets a day
- Flagging duplicate listings in a product catalogue
- Replacing a keyword filter that mis-tags half its matches
Poor fits
- Drafting customer emails or release notes
- Reviewing a few high-stakes contracts a month
- Anything that needs a written explanation
Implications for Enterprise AI Automation
The introduction of Jev signals a potential paradigm shift in enterprise AI, moving away from language models that generate text to models optimized for decision-making within software. By providing structured, probabilistic answers that can be directly acted upon, Jev could significantly reduce costs, improve speed, and enhance reliability in automating routine judgments—such as support tickets, risk assessments, or compliance checks. This approach addresses longstanding issues with LLMs, like hallucinations and overconfidence, which have limited their use in critical decision contexts.
For organizations, this could mean a broader adoption of AI in internal workflows, where the focus shifts from human-like conversation to precise, schema-bound decisions. If proven reliable, System One Models like Jev could replace many current reasoning models, leading to faster, more consistent automation and reducing the need for human oversight in routine tasks.
enterprise decision automation AI tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Evolution of AI Models and Enterprise Adoption
Over the past three years, the AI industry has seen a proliferation of large language models promising improved reasoning, longer context windows, and better code generation. Companies like OpenAI, Anthropic, and Google have heavily invested in chat-based models aimed at human-like interaction. However, these models often struggle with reliability, hallucinations, and high costs, especially when deployed at scale inside enterprises.
TypeSafe’s approach with Jev and System One Models marks a departure from this trend, emphasizing decision accuracy over free-form language. Diogo Almeida, who helped develop RLHF techniques for chatbots, now advocates for a training method—RLCD—that aims to produce calibrated, dependable decisions. The launch of Jev with $40 million in funding led by DCVC underscores investor interest in this new direction. While traditional models remain dominant, early benchmarks and internal tests suggest System One approaches could complement or replace parts of existing AI workflows, especially where speed and cost are critical.
“Jev is designed to produce decisions, not words. It behaves more like a function call than a chat partner.”
— Diogo Almeida, co-inventor of RLHF
structured decision-making AI software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Accuracy and Reliability of System One Models
While Jev shows promise in speed and cost, its accuracy varies across tasks. Benchmarks indicate it performs well on narrow, well-defined questions but less reliably on broader or ambiguous ones. The model’s overconfidence in some cases and underconfidence in others suggest that, like all AI models, it requires careful calibration and validation before widespread deployment. Independent tests have yet to fully validate its performance outside controlled benchmarks, and real-world effectiveness remains to be seen.
probabilistic decision models for enterprise
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Upcoming Validation and Industry Adoption Trials
TypeSafe plans to expand testing of Jev within different enterprise environments, focusing on automating routine decisions in customer support, compliance, and operational workflows. Further independent evaluations are expected to clarify its accuracy and reliability in diverse settings. The company also intends to refine training techniques and improve calibration, aiming for broader adoption of System One Models as a complement or alternative to traditional language models in enterprise AI.
As an affiliate, we earn on qualifying purchases.
Key Questions
How does Jev differ from traditional language models?
Jev produces structured, typed decisions with associated probabilities, rather than free-form text. It behaves more like a function call within software, enabling direct automation of decisions.
Can Jev replace human judgment entirely?
While Jev aims to improve automation speed and reliability, its accuracy varies depending on the task. It is intended to assist, not fully replace, human oversight until proven consistently reliable across use cases.
What are the main limitations of Jev currently?
Its accuracy on complex or ambiguous questions is still being evaluated. Benchmarks show variable performance, and overconfidence or underconfidence in probability estimates remains a concern.
Will this approach work for all enterprise AI applications?
Jev is designed for specific decision-making tasks where structured responses are preferable. Its effectiveness outside these contexts is still under investigation, and it may complement rather than replace existing models.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
