--- title: Jev description: "TypeSafe AI's model family for typed decisions and probabilities from supplied text, introduced as its first System One model." canonical_url: 'https://darkfactory.dev/glossary/jev' markdown_url: 'https://darkfactory.dev/glossary/jev.md' collection: glossary date_published: '2026-10-07T00:00:00-04:00' date_modified: '2026-10-07T00:00:00-04:00' --- # Jev ## Definition Jev is TypeSafe AI's family of models that evaluate supplied text and return typed decisions with probabilities. It is the company's first public System One model. Applications define the available choices or scoring criteria, supply relevant state, and consume the results in code. Jev's current question types are Choice, Score, and Noul, a probability for a true-or-false condition. These primitives support narrow judgments such as routing a ticket or evaluating whether a message meets a stated condition. A Jev call supplies predictions; the application's policy determines their consequences. ## Origin and attribution TypeSafe released Jev in early access on September 15, 2026. Diogo Almeida, TypeSafe's cofounder and CEO, wrote the launch announcement and explained the product in his September 21 interview. Credit for developing the model belongs to the TypeSafe team. The company named Jev after economist William Stanley Jevons. Almeida presents useful intelligence per dollar as the family's design target. ## Family, versions, and capabilities The October 7, 2026 model catalog lists Jev 1.13 with the versioned API ID `jev-1.13.0`. The aliases `jev-latest` and `jev-preview` currently select that release, but the provider can move them to later versions. Pin an evaluated ID when a workflow needs reproducible version selection, and log the returned model ID. That catalog describes text input, including strings and structured text data, without native image, audio, or video input. It also says customer requests shape behavior through state and question criteria, rather than account-specific fine-tuning or LoRA weights. TypeSafe describes Jev's training as reinforcement learning for calibrated decisions (RLCD). ## Limits and disputed claims The launch says Jev cannot hallucinate; its supporting discussion establishes schema matching. A permitted answer can still be wrong. TypeSafe's Jev 1.13 documentation records problems with numerical precision, date comparisons, indirection, adversarial content, and choice ordering. It recommends performing arithmetic and date comparisons in code. Its published speed and cost comparisons describe TypeSafe's selected workflows and comparison conditions. They do not establish a universal advantage across tasks or deployments. Evaluate the complete workflow, including input preparation, retries, and the cost of errors. Choice and Score return a probability distribution and a derived confidence statistic. Noul returns a probability without that confidence field. Neither typed output nor a concentrated probability distribution establishes that a particular answer is true. ## Distinguish it from nearby terms - System One names TypeSafe's proposed model class. Jev names a concrete family within that proposal. - A decision model is a general description of bounded judgment output. Jev is one provider's implementation. - Structured output constrains format. Jev's calibration and accuracy need their own evaluation. - Model routing can use a Jev judgment to choose a downstream model. A model cascade can escalate difficult cases; neither pattern is built merely by selecting the Jev family. ## Check your understanding A workflow tested against `jev-latest` starts producing different routing decisions after an update. Which records would show whether the model version changed, and what would you re-evaluate before retaining the existing confidence thresholds? ## Also called Jev model ## Related terms - [System One](https://darkfactory.dev/glossary/system-one) - [Decision model](https://darkfactory.dev/glossary/decision-model) - [Classification](https://darkfactory.dev/glossary/classification) - [Calibration](https://darkfactory.dev/glossary/calibration) - [Structured output](https://darkfactory.dev/glossary/structured-output) - [Model routing](https://darkfactory.dev/glossary/model-routing) - [Model cascade](https://darkfactory.dev/glossary/model-cascade) - [Context window](https://darkfactory.dev/glossary/context-window) - [Evaluation (eval)](https://darkfactory.dev/glossary/evaluation) - [Useful intelligence per dollar](https://darkfactory.dev/glossary/useful-intelligence-per-dollar) ## Related factory areas - [Model selection, routing & budgets](https://darkfactory.dev/factory/model-routing-budgets) - [Tools & project interfaces](https://darkfactory.dev/factory/tools-interfaces) - [Verification, evaluation & quality truth](https://darkfactory.dev/factory/verification) ## Evidence and further reading - [TypeSafe: Introducing System One Models & Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) - [TypeSafe: System One](https://docs.typesafe.ai/concepts/system-one) - [TypeSafe: Jev 1.13 jaggedness](https://docs.typesafe.ai/model-jaggedness/jev-1.13) - [Latent Space: Jev with Diogo Almeida](https://www.latent.space/p/jev) - [TypeSafe: Models](https://docs.typesafe.ai/models) - [TypeSafe: AI primer](https://docs.typesafe.ai/introduction/machine-learning-primer) - [TypeSafe: Primitives](https://docs.typesafe.ai/primitives) - [TypeSafe: Confidence](https://docs.typesafe.ai/confidence) - [TypeSafe: Team](https://typesafe.ai/team)