Amazon is building its own artificial intelligence model that integrates advanced "reasoning" capabilities. The product is tentatively scheduled to launch in June under the Nova brand, a set of generative artificial intelligence models Amazon released late last year, according to a person directly involved in the project. The person spoke on condition of anonymity because they were not authorized to speak to the media.
The person added that Amazon hopes the new model will adopt a "hybrid reasoning" approach to provide quick answers and more complex extended thinking in a single system. An Amazon spokesperson did not respond to a request for comment.
Inference models have recently emerged as the next frontier in artificial intelligence. Inferential models typically work slower, but can also solve tougher problems by trying multiple solutions and backtracking using thought chaining techniques. Companies like Google, OpenAI, and Anthropic have all recently released their own inference models, and DeepSeek has attracted a lot of attention for building similar products more efficiently.
One of Amazon's top priorities is to make its Nova inference model more price-efficient than competitors, including OpenAI's o1, Anthropic's Claude 3.7 Sonnet and Google's Gemini 2.0 Flash Thinking, according to people involved in the project.
Amazon has previously said its existing in-house Nova models are at least 75% cheaper than third-party models offered through its Bedrock AI development platform.
Another goal of Amazon's upcoming inference model is to rank in the top five for performance based on external benchmarks that assess software development and math skills, such as SWE, Berkeley Function Call Ranking and AIME, the person added.
The move reflects Amazon's determination to invest in its own line of AI-powered models, even as it touts the need for multiple model options through Bedrock. Amazon's AGI team, headed by chief scientist Rohit Prasad, has been working on the new model.
It also puts Amazon into more direct competition with artificial intelligence startup Anthropic, which just launched its latest model. Claude3.7Sonnet uses a similar hybrid approach, combining quick answers with longer thought chain output.
Amazon has invested $8 billion in Anthropic so far, and the two companies have been close partners, collaborating in areas such as artificial intelligence chips and cloud computing.