Amazon is entering the race for AI reasoning capabilities with a new model expected by June under its "Nova" brand. The company aims to catch up with competitors like OpenAI, Anthropic, and Chinese startup DeepSeek through a combination of competitive pricing and strong benchmark performance.
According to Business Insider, Amazon's approach mirrors what Anthropic recently demonstrated with Claude 3.7 Sonnet - a hybrid system that can provide both quick responses and handle complex reasoning tasks. This strategy makes sense given Amazon's substantial partnership with Anthropic, which includes investments of around eight billion US dollars.
Reasoning models represent the cutting edge of AI development today. While slower than conventional language models, they can tackle more difficult problems by exploring multiple solution paths and making their thought processes transparent to users.
Lower prices with competitive performance
Amazon plans to undercut competitors like OpenAI, Anthropic, and Google on price. The company recently stated that its current Nova models are already at least 75 percent cheaper than third-party models available through Amazon's Bedrock AI development platform.
Performance is also a priority, with Amazon expecting its new reasoning model to rank in the top 5 in external benchmarks for software development and mathematics. While this would put Amazon among the leaders, it suggests the company is still catching up rather than setting new standards in the field.
AWS creates dedicated Agentic AI team
Meanwhile, Amazon's cloud division AWS is forming a new Agentic AI group, according to an internal email obtained by Reuters. These AI systems are designed to proactively complete tasks without explicit user requests - similar to features Amazon announced last week for the latest version of its Alexa voice assistant.
AWS CEO Matt Garman sees Agentic AI as a potential billion-dollar business opportunity, believing AI agents will drive the next wave of innovation. The new group will be led by Swami Sivasubramanian, previously Vice President for AI and Data, who will report directly to Garman.
In a separate internal email, AWS Senior Vice President Peter DeSantis outlined additional restructuring to accelerate innovation, including combining the Bedrock and SageMaker AI groups with hardware engineering in the compute organization.
AWS remains a critical component of Amazon's AI strategy, making this focus logical. In November, Amazon made further investments to reduce its dependence on Nvidia graphics cards.
Despite Amazon's ambitious plans, it's worth noting that reasoning capabilities in current AI models remain relatively basic. Even specialized reasoning models like OpenAI's o3-mini (high) achieve only mediocre results in current benchmarks with 44.8 percent accuracy. The development of AI agents is similarly still in its early stages.
