Russian companies overpay up to nine times for access to foreign artificial intelligence models compared to the prices offered by their developers. Domestic providers of such services explain the difference by the need to transfer ruble payments, guarantee the availability of services and ensure legally clear access.
Image source: Solen Feyissa / unsplash.com
For a significant difference in prices noticed administration of the platform for creating AI agents Nodul. The calculations were based on the consumption scenario of 2 million input and 10 million output tokens – this amount of resources is spent on one to three days of work by a specialist with an assistant programmer or a one-time review of an average code base. Foreign tariffs were recalculated at the rate of 82 rubles. for $1; discounts and taxes were not taken into account.
The highest, nine-fold, markup was found on the Yandex AI Studio platform when accessing DeepSeek V4.1 Flash models – 4,000 rubles. versus 443 rub. in the original. Purchasing access to Anthropic Claude Opus 5.5 through a Russian provider costs 14.2 thousand rubles. versus 6.5 thousand rubles. when contacting directly – more than twice as expensive. The difference was similar when accessing the Claude Sonnet 5.5 and OpenAI GPT-6.1 Sol models – 7.1 thousand rubles. versus 3.2 thousand rubles. first hand.
In addition to the actual access to the model via API, the Russian provider takes on international payments, commissions, taxes, document flow and providing access to foreign services – costs are required even for simple proxying of requests if the service must remain stable, ITGlobal.com noted. Yandex AI Studio recalled that they provide access not only to foreign, but also to domestic AI models – on their platform, access to Yandex models is 31% cheaper.
Foreign providers have no obligations to the Russian user, explained Cloud.ru – the platform offers access without VPN, with payment in rubles, guarantees of availability, data protection and integration with other products. In addition, when connecting through a chain of intermediaries, no one can guarantee that requests and responses from models are not intercepted, and the accounts through which access is made will not be suddenly disabled, added to the AI Platform MWS Cloud (MTS).
Another option to ensure stable work with AI is to run models locally. Having analyzed the prices for direct and indirect access to AI models, Ensign purchased workstations for employees and deployed free Chinese models on them – this solution pays for itself in less than a year.
If you notice an error, select it with the mouse and press CTRL+ENTER.
