Manuel Gómez and I, Jorge Mediavilla, mentioned in the latest episode of La Hora de CMS MAG—a special on artificial intelligence (AI) that you can watch for free on YouTube—that Amazon had invested around $8 billion in AI, but had yet to release any related products. That has now changed as Amazon has officially entered the battle for AI supremacy.
At the re:Invent event, Amazon surprised attendees by unveiling six new foundational models, which will be exclusively available on Amazon Bedrock. It’s not a secret that Bedrock is a highly utilized service by various CMS platforms to deliver AI-powered services.
According to the press release, Amazon offers two types of models: a comprehension model that processes text, images, or videos to generate textual output, and a creative content generation model that takes text and image inputs to produce image or video outputs.
Comprehension models
Under the comprehension category, there are three levels, with a fourth, Premier, coming soon:
- Micro: Optimized for speed and lower costs, suitable for tasks like text summarization, translation, content classification, interactive chat, brainstorming, simple mathematical reasoning and basic coding.
- Lite: An ultra-low-cost multimodal model that processes images, videos, and texts to generate text outputs with exceptional speed.
- Pro: A high-capacity multimodal model offering the best combination of precision, speed, and cost for a wide range of tasks, especially those requiring advanced reasoning capabilities. It excels at analyzing financial documents.
- Premier: Set to launch in 2025, this will be the most powerful model, designed for customized use cases.
Content generation models
The creative content generation models will consist of two offerings:
- Canvas: An image generation model capable of producing studio-quality images with precise control over style and content. It includes advanced editing features like image quality enhancement, color correction, and background removal.
- Reel: A more advanced model focused on video generation, enabling the creation of videos from text or images while allowing users to control visual style and pacing.
These models will support up to 300,000 context tokens, handle more than 200 languages, and deliver exceptional performance powered by Llama 3.
Initially, the models will only be available in the United States and will feature security measures and watermarking, though further details about these features are not yet clear.
Pricing
Based on information shared on José Luis Hernández’s LinkedIn, the pricing per million tokens is as follows:
- Micro: $0.035 input / $0.14 output
- Lite: $0.06 input / $0.24 output
- Pro: $0.80 input / $3.20 output
Note: Article originally written in Spanish, translated with ChatGPT, and reviewed in english by Jorge Mediavilla.
Leave a Reply