According to reports, Amazon is developing a multimodal large language model named Olympus, slated for release as early as next week. Internally referred to as Olympus, this algorithm is capable of processing diverse data types, including text, images, and videos. The model is said to enable users to search for specific video segments using natural language prompts and assist energy companies in analyzing geological data.

Sources indicate that Amazon may showcase the Olympus model for the first time at the AWS re:Invent conference, suggesting that Amazon plans to offer this model through its cloud division, Amazon Web Services (AWS), potentially via the AWS Bedrock service. Bedrock is a managed platform that provides access to advanced models, currently offering over half a dozen models developed by Amazon. Among these, the most advanced is Amazon Titan Text Premier, which can handle prompts of up to 32,000 tokens and is capable of generating text and code, as well as performing complex task decomposition.

Furthermore, Bedrock includes three Amazon-developed models designed to generate embedded data structures for machine learning applications, one of which can handle multimodal data. This feature may enable customers to take advantage of Olympus’s multimodal capabilities.

In addition to Amazon’s proprietary models, Bedrock offers large language models developed by other companies, including those from Anthropic PBC, which has received investment from Amazon. Recently, Amazon increased its investment in Anthropic by $4 billion, as part of a total $8 billion investment.

It is believed that the introduction of the Olympus model may be Amazon’s strategy to reduce its reliance on Anthropic. Other tech giants are also working to internalize more of their AI technology stacks. For instance, Meta Platforms is developing a search engine to lessen its dependence on Microsoft and Google’s search technologies for its chatbots.

Amazon’s AI strategy encompasses both software and hardware. The company has developed two chip product lines, AWS Trainium and AWS Inferentia, optimized for training and inference workloads respectively. Last week, Anthropic announced a collaboration with Amazon to further enhance the performance of the Trainium processor series.