Snowflake (NYSE: SNOW), the Data Cloud company, has unveiled Snowflake Arctic, a next-generation large language model (LLM) designed exclusively to be the most open enterprise LLM on the market. With its unique Mixture of Experts (MoE) architecture, Arctic offers cutting-edge intelligence with unparalleled efficiency at scale. It is optimized for complex enterprise workloads, surpassing various industry benchmarks in SQL code generation, instruction tracking, among others. Additionally, Snowflake releases the weights of Arctic under an Apache 2.0 license and the research details that led to how it was trained, setting a new standard of openness for enterprise AI technology. The Snowflake Arctic LLM is part of the Snowflake Arctic model family, a model family created by Snowflake that also includes the best practical embedded text models for retrieval use cases.
“This is a pivotal moment for Snowflake, with our AI research team innovating at the forefront of AI,” says Sridhar Ramaswamy, CEO of Snowflake. “By offering industry-leading intelligence and efficiency in a truly open manner to the AI community, we are pushing the boundaries of what open-source AI can achieve. Our research with Arctic will significantly enhance our ability to deliver reliable and efficient AI to our customers.”
Arctic marks a milestone with a truly open and widely available collaboration
According to a recent Forrester report, approximately 46% of AI decision-makers in companies worldwide are leveraging existing open-source LLMs to adopt generative AI as part of their organization’s AI strategy. With foundational data from Snowflake for over 9,400 companies and organizations worldwide, all users are being empowered to leverage their data with leading open-source LLMs while offering flexibility and choice regarding the models they work with.
Now, with the launch of Arctic, Snowflake provides a powerful and truly open model with an Apache 2.0 license that allows personal, commercial, and research use without restrictions. Additionally, Snowflake provides code templates and flexible inference and training options so users can quickly start deploying and customizing Arctic using their preferred working environments. These include NVIDIA NIM with NVIDIA TensorRT-LLM, vLLM, and Hugging Face. For immediate use, Arctic is available for serverless inference on Snowflake Cortex, Snowflake’s fully managed service that provides machine learning and AI solutions in the Data Cloud. It will also be available on Amazon Web Services (AWS), along with other model files and catalogs, which will include Hugging Face, Lamini, Microsoft Azure, the NVIDIA API catalog, Perplexity, Together AI, among others.
Arctic provides cutting-edge intelligence with high resource efficiency
Snowflake’s AI research team, composed of a unique selection of industry-leading researchers and systems engineers, took less than three months and spent approximately one-eighth of the cost of training similar models to build Arctic. Trained using Amazon Elastic Compute Cloud (Amazon EC2) P5 instances, Snowflake is setting a new benchmark for how fast advanced open-source enterprise models can be trained, ultimately enabling users to create cost-effective custom models at scale.
As part of this strategic effort, Arctic’s differentiated MoE design enhances both training systems and model performance, with meticulously designed data composition focused on enterprise needs. Arctic also offers high-quality outputs, activating 17 out of the 480 billion parameters at a time to achieve industry-leading quality with unprecedented symbolic efficiency. In an efficiency breakthrough, Arctic activates approximately 50% fewer parameters than DBRX, and 75% fewer than Llama 3 70B during inference or training. Additionally, it outperforms leading open models like DBRX, Mixtral-8x7B, and others in encoding (HumanEval+, MBPP+) and SQL generation (Spider), while providing leading language comprehension performance (MMLU).
Snowflake continues to accelerate AI innovation for all users
Snowflake continues to provide companies with the foundational data and cutting-edge AI building blocks they need to create powerful AI and machine learning applications with their enterprise data. When accessed through Snowflake Cortex, Arctic will accelerate customers’ ability to create production-level AI applications at scale, within the security and governance perimeter of the Data Cloud.
In addition to the Arctic LLM, the Snowflake Arctic model family also includes the recently announced integrated Arctic, a next-generation embedded text model family available to the open-source community under an Apache 2.0 license. The family of five models is available on Hugging Face for immediate use and will soon be available as part of the Snowflake Cortex embedded feature (in private preview). These integration models are optimized to deliver leading retrieval performance with approximately one-third of the size of comparable models, providing organizations with a powerful and cost-effective solution for combining their own datasets with LLMs as part of augmented retrieval or semantic search service.
Snowflake also prioritizes giving customers access to the newest and most powerful LLMs in the Data Cloud, including the recent additions of Reka and Mistral AI models. Additionally, Snowflake recently announced expanding its partnership with NVIDIA to continue its AI innovation, joining NVIDIA’s full-stack accelerated platform with Snowflake’s Data Cloud to deliver a secure and exceptional combination of infrastructure and computing capabilities to unlock AI productivity. Snowflake Ventures has also recently invested in Landing AI, Mistral AI, Reka, and other companies to further Snowflake’s commitment to helping customers create value from their enterprise data with LLMs and AI.
Expert AI Comments on the Announcement
“Snowflake Arctic is poised to drive significant outcomes that expand our strategic partnership, driving access to AI, democratization, and innovation for all,” says Yoav Shoham, Co-Founder and Co-CEO of AI21 Labs. “We are excited to see Snowflake helping companies harness the power of open-source models, as we have done with our recent launch of Jamba, the first Mamba-based Transformer-SSM production model. Snowflake’s continued investment in AI is a key factor in our choice to build on the Data Cloud, and we look forward to continuing to create greater value for our joint customers.”
“Snowflake and AWS are committed to generative AI transforming virtually all customer experiences we know,” says David Brown, Vice President of Compute and Networking at AWS. “With AWS, Snowflake was able to tailor its infrastructure to accelerate time to market for training Snowflake Arctic. By using Amazon EC2 P5 instances with Snowflake’s efficient training system and model architecture co-design, Snowflake was able to quickly develop and deliver a new enterprise model to customers. And with plans for Snowflake Arctic to be available on AWS, customers will have more options to leverage powerful AI technology to accelerate their transformation.”
“As the pace of AI continues to accelerate, Snowflake has established itself as an AI innovator with the release of Snowflake Arctic,” says Shishir Mehrotra, Co-Founder and CEO of Coda. “Our innovation and design principles align with Snowflake’s forward-looking approach to AI and beyond, and we are excited to be a partner on this transformation journey of everyday applications and workflows through AI.”
“In recent months, there has been a massive surge in open-source AI,” says Clement Delangue, CEO, and Co-Founder of Hugging Face. “We are excited to see Snowflake significantly contribute to this release not only of the model with an Apache 2.0 license but also with details on how it was trained. It provides the transparency and control necessary for companies to build AI and for the field as a whole to forge new paths.”
“Lamini’s vision is to democratize AI, empowering everyone to build their own superintelligence. We believe the future of enterprise AI is to build on the foundations of powerful open models and open collaboration,” says Sharon Zhou, Co-Founder, and CEO of Lamini. “Snowflake Arctic is important in supporting that AI future. We are excited to fine-tune and customize Arctic for highly accurate LLMs, optimizing control, security, and resilience to a dynamic AI ecosystem.”
“Community contributions are key to unlocking AI innovation and creating value for all,” emphasizes Andrew Ng, CEO of Landing AI. “Snowflake’s open-source release of Arctic is an exciting step in making cutting-edge models available for everyone to fine-tune, evaluate, and innovate on.”
“We are delighted to expand enterprise customers’ options in the fast-evolving AI landscape, bringing the robust capabilities of Snowflake’s new LLM Arctic model to the Microsoft Azure AI model catalog,” says Eric Boyd, Corporate Vice President of Azure AI Platform at Microsoft. “Our collaboration with Snowflake is an example of our commitment to driving open innovation and pushing the boundaries of what AI can achieve.”
“The ongoing advancement – and healthy competition among – open-source AI models is fundamental not only to Perplexity’s success but to the future democratization of generative AI for all,” notes Aravind Srinivas, Co-Founder, and CEO of Perplexity. “We look forward to experimenting with Snowflake Arctic to customize it for our product, ultimately generating even greater value for our end users.”
“Snowflake and Reka are committed to putting AI in the hands of every user, regardless of their technical expertise, to drive business outcomes faster,” says Dani Yogatama, Co-Founder, and CEO of Reka. “With the launch of Snowflake Arctic, Snowflake is advancing this vision by making truly open global large language models accessible to users.”
“As an organization at the forefront of open-source AI research, models, and datasets, we are thrilled to witness the launch of Snowflake Arctic,” says Vipul Ved Prakash, Co-Founder, and CEO of Together AI. “Advancements in the open-source AI landscape benefit the entire ecosystem and enable developers and researchers worldwide to deploy impactful generative AI models.”