t its annual re:Invent conference in Las Vegas, Amazon Web Services used its keynote stage to make a series of generative AI announcements that collectively represent the most aggressive product push in the platform's history.

Nova Tiers3Micro, Lite, Pro
Claim vs GPT-4oLower costOn select internal benchmarks
Bedrock partners100+Model providers on platform

The Amazon Nova Model Family

Amazon Nova, the in-house model suite unveiled at the conference, spans three tiers: Nova Micro for low-latency text tasks, Nova Lite for multimodal use cases, and Nova Pro, the flagship model targeting complex reasoning and long-context applications. AWS claims Nova Pro outperforms GPT-4o on several internal benchmarks at a significantly lower inference cost, though independent verification is pending.

Bedrock Agents V2

The second major announcement was Amazon Bedrock Agents v2, which introduces a visual workflow builder for orchestrating multi-agent pipelines without writing coordination code. The feature is an explicit response to Microsoft's AutoGen and OpenAI's Assistants API.

AI is not a feature. It is a fundamental platform shift — the same as the cloud itself.

Andy Jassy, CEO, Amazon

CEO Andy Jassy used his keynote to frame AI not as a feature but as a fundamental platform shift, drawing parallels to the company's early commitment to cloud infrastructure when the market was sceptical. AWS currently holds the largest share of cloud infrastructure spend globally, and its AI push is aimed at ensuring that lead persists into the model-serving era.