A former OpenAI researcher has released a fundamentally different kind of AI model that abandons language entirely, instead outputting probabilities and decisions that developers say are faster, cheaper, and more reliable than ChatGPT for automation tasks. The model, called Jev, represents a sharp departure from the large language model (LLM) approach that has dominated AI development since ChatGPT's launch in 2022.

Why Did an OpenAI Researcher Leave to Build Something Completely Different?

Jev was created by Almeida, an OpenAI researcher who helped build ChatGPT and invented reinforcement learning from human feedback (RLHF), the training technique largely responsible for making modern AI systems conversational and useful. Despite these accomplishments, Almeida grew frustrated with the limitations of language-based AI.

"We have lightning in a bottle, and yet it is not useful," Almeida told TechCrunch. "I've been battling that problem since then. It took me a while to come to the conclusion: the problem is we are optimizing for human language. We have been super good at human language for four years, but it's not useful for automation because computers speak a different language."

Almeida, Researcher at TypeSafe AI

Two years ago, Almeida left OpenAI to start TypeSafe AI, a startup focused on solving this exact problem. This week, the company released Jev, a transformer-based model that fundamentally rethinks how AI should work for software automation.

How Does Jev Work Differently Than ChatGPT?

Instead of generating text like ChatGPT does, Jev produces probabilities and what the company calls "calibrated decisions." This seemingly small difference creates major practical advantages. Because users define the outputs in advance, the model cannot hallucinate or make up information. Its output tokens are free, and input tokens are metered by the billion rather than the million, making it dramatically cheaper to operate.

The model is trained exclusively on synthetic data using a technique Almeida calls "reinforcement learning from calibrated decisions." Almeida emphasized the importance of this approach, stating that the decision to create all of their training data internally has been one of the best strategic bets the company has made.

What Real-World Performance Gains Are Developers Seeing?

Early adopters report significant improvements in speed and cost. Pranit Sharma, a software engineer at Vercel, a company building agentic infrastructure, tested Jev against OpenAI's ChatGPT Luna 5.6 for running a classifier to review commands for safety. When Vercel switched from ChatGPT Luna to Jev, the results came back 5 to 18 times more quickly and with greater accuracy.

In another test, Bryo AI's CTO Nikhil Mudholkar compared Jev against Google's Gemini for classifying business emails. While Gemini was slightly more accurate, it was 10 to 20 times more expensive. More importantly to Mudholkar, Jev provided real probability scores that make it ideal for automating workflows.

Steps to Understand Jev's Practical Applications

• Software Automation: Jev is most useful for incorporating intelligence into code without the hallucination risks and high costs of language models, making it ideal for developers building automated systems.
• Model Monitoring: Jev can act as a smart check on misbehavior from other AI agents, tracking language model traces and preventing jailbreaks at a fraction of the cost of using another LLM for oversight.
• Model Routing: The model can predict whether a given workload requires a specific AI model, enabling real-time sorting and optimization without the expense of using a language model for the decision.

Armin Ronacher, the CTO of Earendil, which builds the open-source model harness Pi, explained the practical implications of Jev's confidence scores. "At the end of the day, it delegates the hallucination problem a little bit to the user," Ronacher noted. "The user has to say, okay, if this only comes back with 50% probability, maybe this is a coin toss, and I disregard it. But if it's 95%, sure, then I can do something with it".

What Does This Mean for the Future of AI Models?

Jev's success suggests that the dominance of large language models may not be universal. The model is named after William Stanley Jevons, the 19th-century economist whose eponymous paradox describes how falling costs of a commodity lead to increased usage. Almeida hopes that falling costs of AI intelligence will lead to widespread deployment of smart software throughout systems, much like the early internet rather than today's mega-apps.

Demand for Jev has been so high that TypeSafe AI briefly lost the ability to serve users from its API. Ronacher expects competitors will emerge now that the utility of this approach is apparent. "We should have seen this earlier in many ways, but presumably because the LLMs are so cheap and subsidized, you often don't have to be creative yet," he observed.

When asked whether TypeSafe AI is a frontier lab, Almeida pushed back on the framing. "The main product of Frontier Labs is fear or hype. I would like our main product to be intelligence," he stated. "We are not a lab in the sense of, you know, like bet on infinite wealth, or a religion, or building God in a data center, or whatever is the thing of today".

TypeSafe is already planning to build more versions of Jev in new modalities, suggesting that this non-language approach to AI may expand beyond text classification and automation tasks. For developers tired of paying premium prices for language models that hallucinate, Jev represents a fundamentally different bet on what AI should be optimized for.