Amazon's Most Advanced AI Model, Nova Premier, Debuts

Amazon has unveiled what it describes as the most advanced model in its Nova AI family: Nova Premier.
Available on Amazon's AI development platform, Bedrock, Nova Premier processes text, images, and video (though not audio). Amazon states that the model excels at complex tasks demanding deep contextual understanding, multi‑step planning, and precise execution across various tools and data sources.
The Nova lineup was first introduced last December at AWS re:Invent. Since then, Amazon has expanded the family with models capable of generating images and videos, understanding audio, and performing autonomous tasks.
With a context length of 1 million tokens—enabling it to analyze roughly 750,000 words at once—Nova Premier still falls short on certain benchmarks compared to rival flagship models. On the SWE‑Bench Verified coding test, for instance, it trails Google's Gemini 2.5 Pro. It also shows weaker results on GPQA Diamond and AIME 2025, which measure math and science proficiency.
However, Amazon’s internal benchmarking highlights Premier's strengths in knowledge retrieval and visual understanding, as seen in its performance on SimpleQA and MMMU tests.
Pricing in Bedrock is set at $2.50 per million input tokens and $12.50 per million output tokens, placing it roughly in line with Gemini 2.5 Pro, which costs $2.50 per million input and $15 per million output tokens.
Notably, Premier is not a "reasoning" model. Unlike OpenAI’s o4‑mini or DeepSeek’s R1, it does not take extra time or computing resources to deliberate and fact‑check its responses.
Techcrunch event Exhibit at TechCrunch Sessions: AI
Reserve your table at TC Sessions: AI and showcase your solutions to 1,200+ decision‑makers—without a major investment. Offer valid through May 9 or until space runs out.
Exhibit at TechCrunch Sessions: AI
Reserve your table at TC Sessions: AI and showcase your solutions to 1,200+ decision‑makers—without a major investment. Offer valid through May 9 or until space runs out.
Berkeley, CA | June 5 BOOK NOW Amazon positions Premier as especially suitable for distilling knowledge into smaller models—effectively transferring its specialized capabilities into faster, more efficient packages for specific use cases.
AI is becoming increasingly central to Amazon's overall growth strategy. CEO Andy Jassy recently noted that the company is building over 1,000 generative AI applications, with AI revenue growing at triple‑digit annual rates and reaching a multi‑billion‑dollar annual run rate.
Related article
Amazon introduces Alexa for Shopping while pushing Rufus to the background
Amazon has launched Alexa for Shopping, merging its Rufus shopping chatbot with Alexa+ across the app, website, and Echo Show devices.The assistant answers product queries, compares items, tracks prices, and supports shopping reminders. It also handl
Does Mark Zuckerberg Really Believe AI Is for Everyone?
Loading the player…Meta introduced Glimmer this week, an open-weight AI model that anyone can download and run on personal hardware—a sharp contrast to Muse Spark, the company’s more powerful model, which remains locked behind its own APIs. The relea
TechCrunch Disrupt 2026 to showcase leaders from Amazon, Replit and Tether
TechCrunch Disrupt has long been the epicenter of technology’s most critical dialogues, with a legacy spanning over a decade. The speakers who have graced this stage are icons of both the startup era and the current AI revolution, including Zuckerber
Related Special Topic Recommendations
Comments (0)
0/500

Amazon has unveiled what it describes as the most advanced model in its Nova AI family: Nova Premier.
Available on Amazon's AI development platform, Bedrock, Nova Premier processes text, images, and video (though not audio). Amazon states that the model excels at complex tasks demanding deep contextual understanding, multi‑step planning, and precise execution across various tools and data sources.
The Nova lineup was first introduced last December at AWS re:Invent. Since then, Amazon has expanded the family with models capable of generating images and videos, understanding audio, and performing autonomous tasks.
With a context length of 1 million tokens—enabling it to analyze roughly 750,000 words at once—Nova Premier still falls short on certain benchmarks compared to rival flagship models. On the SWE‑Bench Verified coding test, for instance, it trails Google's Gemini 2.5 Pro. It also shows weaker results on GPQA Diamond and AIME 2025, which measure math and science proficiency.
However, Amazon’s internal benchmarking highlights Premier's strengths in knowledge retrieval and visual understanding, as seen in its performance on SimpleQA and MMMU tests.
Pricing in Bedrock is set at $2.50 per million input tokens and $12.50 per million output tokens, placing it roughly in line with Gemini 2.5 Pro, which costs $2.50 per million input and $15 per million output tokens.
Notably, Premier is not a "reasoning" model. Unlike OpenAI’s o4‑mini or DeepSeek’s R1, it does not take extra time or computing resources to deliberate and fact‑check its responses.
Techcrunch eventExhibit at TechCrunch Sessions: AI
Reserve your table at TC Sessions: AI and showcase your solutions to 1,200+ decision‑makers—without a major investment. Offer valid through May 9 or until space runs out.
Exhibit at TechCrunch Sessions: AI
Reserve your table at TC Sessions: AI and showcase your solutions to 1,200+ decision‑makers—without a major investment. Offer valid through May 9 or until space runs out.
Berkeley, CA | June 5 BOOK NOWAmazon positions Premier as especially suitable for distilling knowledge into smaller models—effectively transferring its specialized capabilities into faster, more efficient packages for specific use cases.
AI is becoming increasingly central to Amazon's overall growth strategy. CEO Andy Jassy recently noted that the company is building over 1,000 generative AI applications, with AI revenue growing at triple‑digit annual rates and reaching a multi‑billion‑dollar annual run rate.
Does Mark Zuckerberg Really Believe AI Is for Everyone?
Loading the player…Meta introduced Glimmer this week, an open-weight AI model that anyone can download and run on personal hardware—a sharp contrast to Muse Spark, the company’s more powerful model, which remains locked behind its own APIs. The relea
TechCrunch Disrupt 2026 to showcase leaders from Amazon, Replit and Tether
TechCrunch Disrupt has long been the epicenter of technology’s most critical dialogues, with a legacy spanning over a decade. The speakers who have graced this stage are icons of both the startup era and the current AI revolution, including Zuckerber





Home






