Urgent.News

What's breaking now, across thousands of outlets.

AI

A new kind of AI model from a ChatGPT inventor is thrilling developers

Jev, a new kind of AI model, is showing developers a cheaper and faster path to software intelligence.

Almeida, a former OpenAI researcher, co-created reinforcement learning from human feedback (RLHF), the method that greatly advanced AI. Disappointed that OpenAI's chatbot, despite its abilities, was not practical, Almeida left to create TypeSafe AI, aiming to solve that issue. The company unveiled a new model, Jev, which stands apart from the large language models (LLMs) by not generating text; instead, it provides probabilities or "calibrated decisions."

This design makes Jev much cheaper and faster than traditional models, as it avoids hallucinations and meters input tokens in billions rather than millions. Developers have quickly taken notice, finding Jev especially useful for software automation. For instance, Vercel, a company specializing in agentic infrastructure, experienced 5 to 18 times quicker and more accurate results with Jev compared to OpenAI's Luna 5.6.

Another developer, Bryo AI CTO Nikhil Mudholkar, found Jev to be significantly more affordable than Gemini for email classification, despite Gemini being slightly more accurate. Jev's confidence scores, which provide real probabilities, make it ideal for automating workflows. Beyond replacing LLMs, Jev can also enhance them by acting as a smart check on misbehavior.

Almeida sees potential in using Jev to monitor LLM agent traces and prevent jailbreaks, effectively delegating the hallucination problem to the user. The model's low cost and speed also make it suitable for model routing, predicting which workload requires a specific model—a task that would be prohibitively expensive with an LLM.

Almeida hopes that Jev's low cost and speed will lead to the widespread deployment of intelligent software, much like the early internet. Despite the model's potential, Almeida remains tight-lipped about its architecture, which is believed to be built on top of an open-weight LLM. TypeSafe AI plans to develop more versions of Jev in different formats, but other companies are expected to follow suit, given the model's demonstrated utility.

Written by urgent.news from TechCrunch's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at techcrunch.com →

More in AI

Airport: Manage Multiple AI Agent Terminals Happily ✈️

AI coding agents have changed the way I develop software. Instead of one assistant helping with one task at a time, I increasingly found myself running multiple agents in parallel — Claude Code…

  • Airport is a VS Code extension for managing multiple AI coding agents.
  • Provides live status icons and files view for agent monitoring.
  • Enables one-click agent launch and session resume features.

Routing OpenCode Tasks with Jev

After the release of the TypeSafe AI Jev model and its availability in OpenRouter, I decided to immediately test its capabilities and rewrite my minimalist model router in OpenCode.

  • Jev AI model integrated into OpenRouter
  • Tool assesses Coordination, Uncertainty, Consequences
  • Complexity determined by higher Coordination/Uncertainty value

More from Friday 18 September →