Microsoft didn’t hang around. Despite the controversy surrounding DeepSeek, the R1 model is now part of the AI models available to customers of Azure, the Windows publisher’s cloud computing platform. Developers will be able to rent computing power from Microsoft to run R1 in their applications.
Fervour around DeepSeek
This is actually not a surprise: Azure AI Foundry, Microsoft’s AI platform, already offers a considerable range of AI models (the catalog contains more than 1,800). The goal is to attract developers, regardless of their preferred model. R1 could be part of it: according to DeepSeek, access to the API costs $2.19 per million tokens generated (a token representing a unit of text, a part of a word or a symbol that the AI processes to generate language).
Compare to $60 to access o1, the OpenAI model… Knowing that the capabilities of R1 are very close, if not better, than those of o1. Developers will quickly do the math! It would have been strange, or at the very least economically inconsistent, for Microsoft not to rush to the DeepSeek model.
This very low cost is one of R1’s strengths, as is its open source access. Assets that have been shaking up US tech giants for a few days, those who swear by proprietary models running on infrastructures worth several hundred billion dollars.
Nevertheless, this rush to integrate R1 may be surprising in light of the discoveries of OpenAI and its partner Microsoft: DeepSeek is said to have harvested OpenAI and ChatGPT for its models. A technique called "distillation", which is not prohibited in itself, but AI companies generally prohibit the creation of new competing models from their own models. An accusation that is not lacking in bite, knowing that OpenAI has not held back from drawing extensively and without authorization from content that is sometimes protected by copyright.
Source: Microsoft
0 Comments