# Jev (TypeSafe AI)

System One, TypeSafe AI. Native probabilities, via Vercel AI Gateway.

**Jev** is TypeSafe AI's hosted decision model, launched on 2026-09-15 and the first *System One model*: it writes no text. You send a *state* (any text or JSON) with typed questions and get probabilities back in one parallel pass: `noul` answers yes or no with P(yes), `choice` picks one of up to 255 options with a probability for each, and `score` places the state on an ordered rubric of 2 to 10 levels. It is called at `api.typesafe.ai/v1/systemone` or through Vercel AI Gateway as `typesafe-ai/jev`, and costs $0.042 per million input tokens, with output free. The numbers below are Jevals' own measurements against human labels, next to LLMs asked the same questions ([how we score](https://jevals.com/methodology/)).

| Board | Decision Score (95% range) | Rank | Accuracy | Calibration gap (ECE, pts) | Hand-off at 95% | $ per 1k decisions | Speed p95 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| [noul (PubMedQA)](https://jevals.com/noul/) | 69.0 (59.8 to 76.5) | tied for rank 1 of 8 | 91.3% | 5.0 | 86% | $0.029 | 653 ms |
| [choice (Banking77)](https://jevals.com/choice/) | 67.8 (61.3 to 74.5) | tied for rank 2 of 8 | 79.7% | 9.8 | 54% | $0.043 | 693 ms |
| [score (HelpSteer2 helpfulness)](https://jevals.com/score/) | 9.2 (−4.4 to 21.0) | tied for rank 1 of 8 | 41.3% | 19.7 | — | $0.036 | 670 ms |

Decision Score: 100 = perfect, 0 = guessing the label base rates, below 0 = worse than that. Ranks: 1 + the number of rows significantly better (paired item bootstrap, 95%); rows that cannot be told apart share a rank. Hand-off at 95%: the share of decisions a model can take alone, at its own confidence threshold, while staying at least 95% right.

Setup: model id typesafe-ai/jev; served as typesafe-ai/jev; adapter ai-sdk experimental_evaluate ai@7.0.106; reasoning —; temperature —; price as of 2026-09-18.

Compare: [Jev vs Gemini 3.8 Flash](https://jevals.com/compare/jev-vs-gemini-3.8-flash/) · [Jev vs GLM-5.3](https://jevals.com/compare/jev-vs-glm-5.3/) · [Jev vs DeepSeek V4.1 Flash](https://jevals.com/compare/jev-vs-deepseek-v4.1-flash/) · [Jev vs Qwen3.8 Flash](https://jevals.com/compare/jev-vs-qwen3.8-flash/) · [Jev vs Mistral Medium 3.5](https://jevals.com/compare/jev-vs-mistral-medium-3.5/) · [Jev vs Mercury 2.5](https://jevals.com/compare/jev-vs-mercury-2.5/)

Source: https://jevals.com/models/jev/ · release 2026-09-18, suite 0.1.0 · data (CC-BY-4.0): https://jevals.com/data/releases/2026-09-18/board.json · cite as: Jevals (jevals.com), release 2026-09-18, suite 0.1.0. CC-BY-4.0.