Amazon.com Inc. has invested $2.75 billion in Anthropic PBC, a startup with a large language model that can outperform GPT-4 across multiple task categories. 

The companies announced the deal today. Amazon provided the capital as part of a previously announced plan to invest up to $4 billion in Anthropic. Under the same agreement, it led a $1.25 billion funding round for the artificial intelligence developer in September.

Anthropic said at the time that it will invest the capital in foundation model development and AI safety research. The company intends to build its next-generation LLMs using Amazon Web Services Inc. infrastructure. Anthropic will leverage instances powered by the cloud giant’s AWS Trainium and AWS Inferentia chips, which are optimized for AI training and inference, respectively.

The companies’ partnership extends to several other areas as well. Anthropic will help AWS develop future enhancements to Trainium and Inferentia. Amazon, in turn, plans to make the AI developer’s software available to its engineers.

Anthropic’s LLMs are also accessible in Amazon Bedrock. Introduced last April, Bedrock is a managed service through which AWS provides access to internally developed and third-party foundation models. Anthropic’s latest LLM family, the Claude 3 series, rolled out to the service earlier this month.

The flagship model in the Claude 3 series is known as Opus. According to Anthropic, it outperformed OpenAI’s GPT-4 in an internal assessment that compared the models using ten popular AI benchmarks. Those benchmarks included math questions, coding tasks and general knowledge questions spanning various topics.

On Tuesday, Opus also surpassed GPT-4 on a leaderboard called Chatbot Arena that researchers use to compare LLMs. OpenAI’s flagship AI model previously topped the ranking for more than six months.

Anthropic says Opus can outperform several other language models as well, most notably Google LLC’s Gemini Ultra. The search giant is both a competitor and a major investor in the company. 

Shortly after Amazon announced its $4 billion Anthropic investment last year, Google agreed to buy a stake worth as much as $2 billion in the AI developer. The search giant earlier backed two funding rounds worth $750 million for Anthropic. The larger of those rounds, a $450 million investment announced last May, also included the participation of Salesforce Ventures and other investors.

Recent reports suggest Anthropic may raise yet more capital in the near future. Late last year, The Information reported the company was holding talks to raise $750 million at a valuation that could exceed $18 billion. Menlo Ventures was expected to lead the round. 

Image: Anthropic