Highlights:

  • The cloud giant’s AWS Trainium and AWS Inferentia processors, designed for AI training and inference, will power instances used by Anthropic.
  • The LLMs from Anthropic are also available on Amazon Bedrock, a managed service launched in April last year.

Amazon invests USD 2.75 billion in Anthropic PBC, a startup developing a large language model that claims to outperform GPT-4 over several task categories.

The firms announced the deal recently. The funding came from Amazon.com Inc. as a part of its previously disclosed intention to spend up to USD four billion in Anthropic. In September, it spearheaded a USD 1.25 billion fundraising round for the creator of artificial intelligence under the same arrangement.

At the time, Anthropic stated that it would use the funds for AI safety research and the construction of foundation models. The organization plans to use technology from Amazon Web Services Inc. to construct its next-generation LLMs. The cloud giant’s AWS Trainium and AWS Inferentia processors, designed for AI training and inference, will power instances used by Anthropic.

The companies also collaborate in a number of other areas. Anthropic will work with AWS to create Trainium and Inferentia’s upcoming improvements. Amazon intends to provide its engineers with access to the software developed by the AI developer.

The LLMs from Anthropic are also available on Amazon Bedrock, a managed service launched in April last year. It gives users access to both third-party and internally designed foundation models via AWS. The Claude 3 series, Anthropic’s newest LLM family, was introduced to the service earlier last month.

Opus is the flagship model in the Claude 3 series. Anthropic claims that in an internal evaluation, which pitted the models against ten widely used AI benchmarks, it beat OpenAI’s GPT-4. These benchmarks comprised arithmetic problems, coding exercises, and general knowledge inquiries covering a range of subjects.

On a leaderboard called Chatbot Arena, which academics use to compare LLMs, Opus also beat GPT-4 on Tuesday. The leading AI model from OpenAI previously held the top spot for more than six months.

Anthropic claims that Opus can surpass several existing language models, the most notable of which is Google LLC’s Gemini Ultra. The massive search engine is a significant investor in the business and a competitor.

Google agreed to purchase up to a USD two billion stake in the AI firm shortly after Amazon revealed its USD four billion Anthropic investment earlier in the year. Previously, the search giant contributed USD 750 million to Anthropic’s two funding rounds. Salesforce Ventures and other investors took part in the bigger of those rounds, with a USD 450 million investment revealed in May.

According to recent sources, Anthropic may soon raise further funding. As per the reports from the previous year’s end, the company was in talks to fund USD 750 million at a valuation that might go up to USD 18 billion. Menlo Ventures was supposed to spearhead the funding round.