We are excited to bring DeepSeek-R1 to Snowflake Cortex AI! As described by DeepSeek, this model, trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT), can achieve performance comparable to OpenAI-o1 across math, code and reasoning tasks. Based on DeepSeek’s posted benchmarking, DeepSeek-R1 tops the leaderboard among open source models and rivals the most advanced closed source models globally. Customers can now request an early preview of DeepSeek-R1 on Cortex AI.
As part of the private preview, we will focus on providing access inline with our product principles of ease, efficiency and trust.
-
The model is available in private preview for serverless inference for both batch (SQL function) and interactive (Python and REST API). To request access during preview please reach out to your sales team. The model will be available only in the requested account.
-
The model is hosted in the U.S. within the Snowflake service boundary. We do not share data with the model provider.
-
Once the model is generally available, customers can manage access to the model via role-based access control (RBAC). Account admins can restrict access by selecting the models approved per governance policies.
Snowflake Cortex AI
Snowflake Cortex AI is a suite of integrated features and services that include fully-managed LLM inference, fine-tuning, and RAG for structured and unstructured data, to enable customers to quickly analyze unstructured data alongside their structured data, and expedite the building of AI apps. Customers can access industry-leading LLMs, both open source and proprietary and integrate these easily into their workflows and applications. Snowflake embraced the open source ecosystem with the support for multiple LLMs from Meta, Mistral and Snowflake. We believe this open access and collaboration will pave the way for expedited innovation in this space.
DeepSeek-R1
Based on DeepSeek’s GitHub post, they directly applied reinforcement learning (RL) to the base model without relying on supervised fine-tuning (SFT) as a preliminary step. This approach allowed the model to explore chain-of-thought (CoT) for solving complex problems, resulting in the development of DeepSeek-R1-Zero. They further mention that the initial model demonstrated capabilities such as self-verification, reflection and generating long CoTs but encountered challenges such as endless repetition, poor readability and language mixing. To address these issues the DeepSeek team describes how they incorporated cold-start data before RL for enhanced reasoning performance.
