Cohere's New AI Command R+ Launches on SageMaker and Azure, but Not Yet on Bedrock

Earlier this week, Cohere, a Toronto-based startup co-founded by former Google AI researchers, launched its enterprise-ready AI large language model (LLM): Command R+. The announcement highlighted impressive metrics showing that Command R+ surpassed OpenAI's GPT-4, the model powering ChatGPT, in various common AI benchmarking tests.

Command R+ is now accessible on Microsoft’s Azure cloud service and, as of today, on Amazon SageMaker, another major cloud platform.

Pricing Structure

On Azure, the cost to access Command R+ is $0.015 per 1,000 tokens for model outputs and $0.003 per 1,000 tokens for input prompts. Tokens represent words or parts of words, and a great interactive example can be found by open-source developer Simon Willison.

Conversely, on SageMaker, pricing is structured differently, with costs ranging from $82.89 to $146.93 per hour for real-time inference, which supports production deployments.

Microsoft's Commitment to AI Models

Microsoft proudly announced on X (formerly Twitter) that Command R+ was "available first on Microsoft Azure." The company released a detailed blog post highlighting its extensive library of over 1,600 foundation models from a variety of providers, including Databricks, Hugging Face, and OpenAI.

"Command R+ is designed for enterprises looking to leverage their internal data for customized and precise language models," stated John Montgomery, Corporate Vice President of AI Platform at Microsoft. He emphasized that Command R+, in combination with Cohere Embed and Rerank, will deliver top-tier accuracy for sophisticated AI applications reliant on enterprise data.

Intense Competition in Cloud AI Offerings

Microsoft's latest move to include Command R+ follows its investment in offering the proprietary Mistral Large model from the French AI startup of the same name. With ongoing support for OpenAI, Microsoft aims to establish Azure as the top destination for enterprise developers seeking diverse AI model options.

Meanwhile, Amazon is actively expanding its model library on Bedrock, its LLM-focused development platform, having recently added Anthropic’s Claude 3 model family. The availability of Command R+ on AWS SageMaker, despite its absence from Bedrock, represents a significant development in the competitive cloud market.

The push to offer a broader selection of AI models by major cloud providers is indicative of their fierce rivalry. The more models a provider can make available, the better it can cater to various customer needs—akin to a supermarket boasting a wide array of brands.

For Cohere, gaining presence on prominent cloud and AI development platforms is crucial for business growth. This expansion could potentially help widen its customer base, especially as it faces challenges in generating revenue, according to reports. In contrast, OpenAI has reportedly attracted significant enterprise customer growth, reaching around 600,000 subscribers.

Omission from Bedrock

Aidan Gomez, Cohere’s CEO and co-founder, acknowledged the absence of Command R+ from Amazon Bedrock on X, encouraging users to request its addition.

When inquiring about the absence of Command R+ on Amazon Bedrock, an AWS spokesperson stated, “AWS does not disclose product roadmaps; however, we are committed to offering the most popular and advanced foundation models to our customers. Our partnerships with model providers like Cohere are vital to this mission, and we look forward to further collaboration.”