Alice AI LLM

Unveiled September 2026 and branded separately from the YandexGPT line, Alice AI LLM is Yandex's new flagship model family for dialogue, retrieval-augmented generation, and agent workflows. Notably, it is released under the Apache 2.0 license with downloadable weights for commercial and research use — a break from the API-only YandexGPT generations — and can be self-hosted entirely inside the user's own infrastructure with fine-tuning supported. The current hosted model offers a 65,536-token context. Yandex positions it as the technological foundation for the next generations of its products, including the Alice assistant.

Limitations: very new with the base release positioned as a starting point — Yandex itself frames it as most useful to specialists for now, with productization still in progress. Russian/English bilingual focus; other languages are secondary. As with all Yandex models, the July 2026 accusation by Sberbank's CEO that YandexGPT-5 is a fine-tuned Alibaba Qwen (denied by Yandex) hangs over Russian-model provenance discussions generally.

No public benchmark numbers located