Amazon launches new large model series

75% cheaper price! Amazon launches a new series of large models, with self-developed 3-nanometer chips set to be launched next year

E-commerce giant Amazon continues to make efforts in the field of cloud computing, launching a new self-developed AI (artificial intelligence) model and chip.

On December 3rd local time, Amazon announced a series of new AI products at the AWS (Amazon Web Services) annual conference, including the new self-developed large model Amazon Nova series and its first AI training chip Tranium3 using 3-nanometer process nodes. Benoit Dupin, Senior Director of Machine Learning and Artificial Intelligence at Apple, also attended the event and stated that Apple is considering pre training its proprietary models using Amazon AI chips.

The all-new Nova series of large models covers text, images, and videos

At the conference, Amazon launched a new generation of basic models, the Amazon Nova series. Users can use generative AI applications supported by Amazon Nova to understand videos, charts, and documents, or generate videos and other multimedia content. This series includes six models, ranging from plain text model Micro, low-cost multimodal model Lite, high-performance multimodal model Pro, to Premier expected to be launched in the first quarter of 2025, as well as two more advanced models Canvas and Reel whose release dates have not yet been determined.

It is reported that the Nova series can support 200 languages, and the prices of Micro, Lite, and Pro are at least 75% cheaper than the best performing models in their respective intelligent categories on their generative AI service platform Amazon Bedrock. They are also the fastest models in their respective intelligent categories. These models will also be integrated into the Amazon Bedrock platform for users to use through APIs, and support custom fine-tuning and distillation.

In addition to the graphic generation model Canvas and video generation model Reel, Amazon also announced that it will release a "speech to speech" model and an "arbitrary to arbitrary" multimodal model in 2025, which will be able to process text, images, audio, and video as input and output.

At the same time, Amazon announced at the meeting that it will strengthen cooperation with AI startup Anthropic, and the two companies will jointly launch the Claude 3.5 Haiku delayed optimization version, which will increase inference speed by 60%. Amazon has invested a total of $8 billion in Anthropic. Tom Brown, co-founder and Chief Computing Officer of Anthropic, also made an appearance at the conference, announcing the company's launch of Project Rainier, which will use a computing power cluster with hundreds of thousands of Amazon self-developed chips in the future.

The first self-developed chip with a 3-nanometer process is expected to be launched next year

After finishing the big model, we came to the AI chip. AWS CEO Matt Garman announced that its Trainium2 chip is now fully available for rent, supported by the Amazon chip Trainium2