AWS subscribers now have entry to generative AI fashions that rival GPT-4o. On Dec. 3, in the course of the AWS re:Invent occasion held in Las Vegas and on-line, AWS introduced six new mannequin sizes for various use instances within the new Amazon Nova household.

“Inside Amazon, we’ve got about 1,000 generative AI functions in movement, and we’ve had a chicken’s-eye view of what software builders are nonetheless grappling with,” Rohit Prasad, SVP of Amazon Synthetic Common Intelligence, mentioned within the press launch.

“Our new Amazon Nova fashions are supposed to assist with these challenges for inside and exterior builders and supply compelling intelligence and content material era whereas additionally delivering significant progress on latency, cost-effectiveness, customization, Retrieval Augmented Technology (RAG), and agentic capabilities.”

What’s Amazon Nova?

Amazon Nova is a line of generative AI basis fashions out there on AWS’s Amazon Bedrock AI internet hosting service. Organizations can experiment with three dimension choices at the moment:

• Amazon Nova Micro is a text-only mannequin with a fast response time of 210 output tokens per second. Amazon claims it outperforms Meta’s Llama 3.1 8B and Google’s Gemini 1.5 Flash-8B. Nova Micro is meant for functions requiring fast responses at a comparatively low price.
• Amazon Nova Lite is one other small mannequin within the Nova household. Not like Micro, it might probably analyze both picture, video, or textual content inputs. Akin to OpenAI’s GPT-4o mini, Nova Lite is meant for fast summarization and interpretation of charts or video shows. As a result of it might probably perceive photographs on laptop screens and carry out perform calling, Amazon Nova Lite is suitable for some quasi-autonomous chained behaviors used for “AI agent” duties.
• Amazon Nova Professional is the mid-range mannequin. Amazon mentioned it performs quicker, extra precisely, and prices lower than OpenAI’s GPT-4o or Google’s Gemini 1.5 Professional. Nova Professional can interpret textual content or photographs and helps agentic workflows.

As soon as prospects have a Nova mannequin, they will fine-tune it based mostly on their proprietary information.

Along with the dimensions choices, organizations also can choose from a picture era mannequin (Amazon Nova Canvas) and a video mannequin (Amazon Nova Reels). Each of those are supposed to create “studio-quality” content material.

• Nova Canvas creates photographs based mostly on textual content or picture prompts. Amazon notes it consists of security options reminiscent of watermarking and content material guardrails.
• Nova Reels creates six-second movies, with Amazon planning to increase the doable video size to 2 minutes in “the approaching months.”

SEE: AI regulation is ongoing in Australia, with a committee calling for giant fashions from OpenAI, Meta, and Google to depend as “high-risk.”

What’s subsequent?

The fourth mannequin within the Nova line, Nova Premier, won’t be out there till the primary quarter of 2025. Amazon expects Nova Premier to carry multimodal (video, picture, or text-to-text) interpretation and a hefty information library that organizations can use to coach different fashions.

Additionally, Amazon plans so as to add a mannequin that may reply naturally to spoken dialog. They’re additionally engaged on a multimodal-to-multimodal mannequin to interpret and output textual content, photographs, video, or audio.

Whereas it’s but too early to see how Nova will compete with rivals like OpenAI, Google, and Meta, Amazon scored one main associate in SAP, which gives the fashions on its AI Core platform.

roosho is masking AWS re:Invent remotely.