🔍 Read the full analysis: Could 'System One' AI Be The Most Useful Model Yet? on ThorstenMeyerAI.com
Get audio and creator gear 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 announced Jev, a new ‘System One’ AI model designed for decision-making within software. It offers faster, cheaper, and more reliable automation by producing typed decisions with calibrated probabilities, challenging traditional large language models.
TypeSafe announced the release of Jev, a new type of AI model designed specifically for decision-making within enterprise software, marking a significant shift from traditional text-generating large language models. This model, part of their ‘System One’ class, produces structured, typed answers with associated probabilities instead of free-form text, aiming to improve speed, cost-efficiency, and reliability in automation tasks. The development challenges the assumption that all AI problems require large language models, offering a potentially more practical alternative for internal business processes.
Jev is built on a novel training approach called Reinforcement Learning for Calibrated Decisions (RLCD), developed by Diogo Almeida, co-inventor of RLHF and InstructGPT. Unlike traditional models that generate text, Jev handles structured questions—choices, scores, and yes/no probabilities—and returns typed decisions with confidence levels. This approach eliminates issues like hallucinations and formatting errors, which are common in large language models, making it more suitable for automation where precision is critical.
TypeSafe claims Jev can answer in 70 to 500 milliseconds at a cost of approximately $0.042 per million tokens, making it significantly faster and cheaper than existing models. The company states that Jev is 193.6 times faster and 444.6 times cheaper than comparable workflows, based on internal benchmarks. These claims highlight the economic and operational advantages of the model, especially for tasks requiring rapid, repeated decision-making within software systems.
While the model is marketed as having ‘zero hallucinations,’ this claim is narrowly defined. Jev cannot invent off-schema answers or produce malformed data structures, but it can still select the wrong answer among its fixed options. Its accuracy has been benchmarked against frontier models like GPT-6 Astra and Claude Fable 5.1, with results indicating room for improvement and emphasizing that accuracy depends heavily on how questions are structured and split into smaller parts.
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 Automation and AI Reliability
The introduction of Jev represents a potential paradigm shift in enterprise AI, moving away from reliance on large language models toward decision-focused, structured outputs. This shift could lead to more dependable automation, reducing errors caused by hallucinations and formatting issues, and lowering operational costs significantly. If adopted widely, Jev could redefine how businesses implement AI for routine decision-making, enabling faster, more cost-effective workflows with higher confidence in outputs.
Moreover, Jev’s approach challenges the notion that complex reasoning always requires large, resource-intensive models. Instead, it suggests that specialized, schema-conformant models can outperform in specific tasks, especially where speed and reliability are paramount. This could influence future AI development strategies, favoring decision-oriented models over text-centric ones in enterprise applications.
However, the model’s accuracy and robustness are still being tested, and its effectiveness across diverse real-world scenarios remains to be seen. Its success could depend on how well organizations adapt their workflows and question structures to leverage its strengths.
enterprise decision-making AI software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background of AI Model Evolution and Industry Shift
Over the past three years, the AI industry has focused heavily on improving large language models (LLMs) like GPT and Claude, emphasizing better reasoning, longer contexts, and code generation. These models have become the backbone of many enterprise AI applications, but they also face limitations such as hallucinations, overconfidence, and high operational costs.
In response, some industry leaders have begun exploring alternative approaches. Diogo Almeida’s work at OpenAI on RLHF helped shape the current landscape, but his new company, TypeSafe, argues that RLHF techniques are poorly suited for automation tasks. Instead, Jev employs RLCD, a training method designed to produce calibrated, decision-oriented outputs.
The launch of Jev marks a notable departure from the trend of scaling large models, proposing a specialized, schema-based AI that prioritizes speed, reliability, and cost-efficiency. Its development is part of a broader industry shift toward more practical, application-specific AI solutions.
“Jev is designed to produce decisions that software can act on directly, eliminating the need for parsing or interpretation.”
— Diogo Almeida, co-inventor of RLHF
Uncertainties in Accuracy and Real-World Performance
While TypeSafe claims significant speed and cost benefits, independent benchmarks reveal that Jev’s accuracy is currently moderate, with scores around 62.6% on phishing email tests—lower than some frontier models. Its overconfidence on certain question types and underconfidence on others suggest that calibration and reliability still need refinement. The true effectiveness of Jev across diverse, complex enterprise scenarios remains unproven, and its long-term robustness is still under assessment.
Additionally, the benchmarks used are based on agreement with other frontier models rather than definitive correctness, raising questions about how well Jev will perform in critical decision-making contexts. The narrow definition of ‘zero hallucinations’ also means that errors related to judgment and decision quality are still possible.
Next Steps for Adoption and Validation
TypeSafe is expected to release further case studies and real-world pilot programs to demonstrate Jev’s capabilities at scale. Industry adoption will likely depend on how well organizations can structure their questions and workflows to maximize decision accuracy and reliability.
Further independent testing and benchmarking are anticipated to better understand Jev’s limitations and potential. As more enterprises experiment with this decision-focused approach, insights into its practical benefits and challenges will emerge, shaping the future of enterprise AI development.
Meanwhile, competitors may explore similar decision-oriented models, potentially accelerating innovation in this niche of AI technology.
Key Questions
How does Jev differ from traditional large language models?
Jev produces structured, typed decisions with associated probabilities, rather than free-form text. This makes it more suitable for automation, reducing errors caused by hallucinations and formatting issues, and enabling direct software actions based on its outputs.
What are the main advantages of Jev?
Jev offers significantly faster response times (70-500 ms), lower costs ($0.042 per million tokens), and higher reliability for decision-making tasks. Its structured outputs eliminate common errors in text generation, making it better suited for enterprise automation.
Are there any limitations to Jev’s accuracy?
Yes. Benchmarks show moderate accuracy levels, around 62-67% agreement with frontier models, and its calibration varies across question types. Its effectiveness depends on how well questions are structured and the specific use case.
Will Jev replace large language models entirely?
Not immediately. Jev targets specific decision-making tasks where speed, reliability, and structured outputs are critical. Large language models may still be preferred for tasks requiring rich, open-ended text generation.
What is the future outlook for decision-focused AI models like Jev?
As enterprises seek more reliable and cost-effective automation, decision-focused models are likely to grow in importance. Further testing and real-world applications will determine how widely Jev and similar models are adopted.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
