Jev focuses AI on fast business decisions

TypeSafe AI released Jev on September 15, 2026 as a model designed to return predetermined options with probabilities rather than write long responses. In e-commerce support, for example, it can force a choice among return, delivery issue, fraudulent use, or human review required, allowing a company system to route the case without generated explanation.

Jev targets the small operational judgments common inside companies, such as where to send an email, whether a transaction looks suspicious, whether an answer contradicts source material, or whether an application can be approved automatically. TypeSafe AI describes the approach as a System One Model, processing multiple questions in parallel for a single input and returning fixed-format answers with probabilities.

The speed and cost claims come from avoiding free-form text generation. TypeSafe AI says a single process takes 70 to 500 milliseconds and is 40 to 200 times faster than conventional AI in its specific business tests, with input cost at $0.042 per million tokens and output described as too cheap to measure.

External testing published on September 29 used 37 types of datasets to execute 346,009 inquiries for less than $10. Jev performed well in text classification and common-sense problems against Qwen and Gemma in many categories, but results varied by language and weakened on fine-grained classification or ambiguous evaluation criteria, making company-specific thresholds such as 90% or higher important for deciding when to automate.