Yandex Pushes Back on Gref Over Qwen AI Claims

Yandex has publicly rebutted German Gref after the Sber chief said Russian AI efforts are effectively built on Chinese models. Yandex says it does use open-source and open-weight systems, including Chinese ones, but only as one tool within a broader in-house development process.

The exchange followed Gref’s remarks at the Federation Council, where he described Sber as the only company with a truly domestic AI stack and said Yandex fine-tunes Qwen rather than building models from scratch. Yandex did not deny that open solutions from China are part of its stack. But it stressed that this does not replace its own research, model training, engineering, and product integration work.

That distinction matters because mixing internal development with open weights and outside architectures has become standard practice for large language models. Startups and major platforms alike use that approach to ship updates faster.

Yandex has acknowledged this before. The company previously said YandexGPT 5 Pro was based on Qwen 2.5 weights, while Alice AI used Qwen 3. According to the source, those disclosures likely prompted Gref’s sharp framing.

Recommended reading

200+ experts urge action on AI’s economic shock

The dispute is also about positioning. Chinese open-weight model families such as Qwen have become a practical foundation for many products outside China, and the real competitive question is often not where the weights came from, but how much has been built on top of them: fine-tuning, infrastructure, training data, and services.

For Russia’s AI race, the public clash looks less like a technical argument and more like a fight over who gets to claim the label of truly homegrown AI. The next round will likely be decided by new model and service releases, not statements alone.