Archive · 21 Sep 2026
Jev AI Model Targets Fast, Cheap Decisions for AI Agents
Jev is a new AI model made by TypeSafe AI.
Unlike a chatbot, it usually does not write paragraphs of text.
Instead, it helps computer programs choose what to do next.
For example, it can decide which tool an AI agent should use or whether the agent should try again.
TypeSafe says Jev gives answers with confidence scores to show how certain it is.
The company says the model is very fast and inexpensive to operate.
Developers are testing it for safety checks, email classification, and routing tasks to other models.
Some early users reported that Jev was faster or cheaper than models from OpenAI and Google, although results differed by task.
TypeSafe AI says Jev could become a decision-making layer beneath many AI agents.
TypeSafe AI launched Jev on September 15 as a model for calibrated decisions rather than text generation.
Jev reportedly answers tool-selection and retry questions in under half a second at $0.042 per million input tokens, with free output.
The model became available without a waitlist on September 21, with access beginning at $5 in credits.
Developers are testing Jev for automation, safety classification, model routing, monitoring, and jailbreak prevention.
TypeSafe AI says Jev uses synthetic data and reinforcement learning for calibrated decisions, while its architecture remains undisclosed.
- Who
- TypeSafe AI, co-founders including Diogo Almeida, and developers testing the Jev model.
- What
- The launch and wider availability of Jev, a transformer-based AI model designed to produce calibrated decisions and probabilities instead of ordinary text.
- When
- Jev launched on September 15; TypeSafe AI announced open availability on September 21.
- Where
- Jev is available through TypeSafe AI’s API and has been added to platforms including Vercel, Cloudflare, LangChain, and Langfuse.
- Why
- The model is intended to help AI agents make fast, lower-cost decisions about tool use, retries, routing, safety, and monitoring.
- Developer
- TypeSafe AI
- Launch date
- September 15
- Open availability
- Announced September 21, without a waitlist
- Input price
- $0.042 per million input tokens
- Output price
- Free
- Response speed
- Under half a second for certain agent decisions
- Training approach
- Synthetic data and Reinforcement Learning for Calibrated Decisions