Yandex's Open-Source Model Among the World's Top AI Releases of the Week

Turing Post highlighted in the release the training from scratch, the open Apache 2.0 license, the hybrid architecture, and the model's results on mathematical and programming tests. Experts linked the emergence of the model to one of the industry's main trends: the industry is moving away from simply increasing model sizes toward growing their efficiency.

With its compact size, the new open-source model from Yandex does not require large computational resources and outperforms many global counterparts in answer quality.

Expert Assessments

Aman Chadha — head of the AI division at Google DeepMind, previously worked at Apple and NVIDIA — placed Alice AI Foundation LLM alongside developments from Google, DeepSeek, Qwen, Z.ai, and Xiaomi. In his publication, Chadha called Yandex's model an example of a model that, with a small number of active parameters, competes with heavier developments.

In his opinion, such efficiency is especially important for AI agents: to complete a single task, an agent may need many calls to the model and external tools.

Founder of Sypha AI Shaun Chauhan called Alice AI Foundation LLM the next serious open base model. He linked its emergence to the return of Russian developers to the global open-source AI agenda.

First Third-Party Developments

Foreign developers have already begun creating their own tools on top of the new model. One Japanese developer adapted Alice AI Foundation LLM for local running on Apple devices. He then prepared an experimental GGUF version and an implementation of the necessary components for llama.cpp.

Alice AI Foundation LLM is not the only open release from Yandex this autumn. In early September, the company published a pretrain version of the Alice AI Search model. It is on its basis that quick answers in Yandex Search operate.