Developers introduced the minojev model for decision making

The development team has released minojev — a specialized model focused on decision-making tasks. Unlike standard language models, the system generates not text answers, but probabilistic distributions, which allows it to be used for automating logical processes.

The architecture of minojev is optimized for single-pass inference. This provides the ability to work in a fully autonomous mode on a regular CPU without the need to connect to cloud servers or specialized GPU accelerators.

The model demonstrates high efficiency in tasks of routing requests to the support service and managing the behavior of autonomous agents. By abandoning text generation, the system minimizes delays, which is critical for scenarios that require an instant response in real-time.

The project's source code is published in open access on GitHub. The developers position the solution as a tool for integration into existing pipelines, where predictable decision-making logic is required without the redundant computational costs characteristic of heavy LLM.

The emergence of such narrowly specialized models makes it possible to transfer part of the intellectual tasks from powerful servers to local devices. This simplifies the implementation of AI logic in systems with limited resources, while maintaining high accuracy of classification and action selection.