Jev

Imagine a ticketing system: whenever a new ticket is submitted, the program must determine which team should be assigned to handle it, what its urgency level is, and whether manual review is necessary. Using large-scale chat models for such tasks requires writing specific prompts, analyzing free-text inputs, and constantly worrying about potential issues with the output format. Jev was precisely designed for scenarios like these.

It is the first model launched by TypeSafe AI, and it is positioned as a “decision-making model” rather than a chatbot. When you provide it with application states and typed questions, it will return options within a specified range, graded scores, or probabilistic outcomes, along with calibrated confidence levels. Developers can set thresholds; if the confidence level is high enough, the program will automatically execute the relevant actions. If not, the task will be handed over to human operators. Additionally, multiple decision-making processes can be combined into a complete workflow within the code, encompassing tasks such as routing, scoring, information extraction, and verification.

The price officially announced on the website is $42 per billion input tokens. Both the speed and cost advantages are figures provided by the manufacturer itself; currently, there are no third-party benchmarks available. This service is still in the early access phase and can be used through APIs and consoles. It is recommended to conduct a small-scale comparison using your own data before integrating it into your systems.