Amazon / AWS AI Models — Complete Guide
🟠 Amazon / AWS AI Models — Complete Guide
Section titled “🟠 Amazon / AWS AI Models — Complete Guide”Amazon’s first-party AI models are delivered through Amazon Bedrock under the Amazon Nova brand. Nova is Amazon’s current-generation model family spanning text reasoning, multimodal understanding, speech, image generation, and video. The older Amazon Titan series remains available for legacy workloads.
🗺️ Amazon AI Model Families
Section titled “🗺️ Amazon AI Model Families”| 🗂️ Family | 🎯 Purpose | 🌟 Status |
|---|---|---|
| 🚀 Amazon Nova 2 | Next-gen reasoning models with 1M context | ✅ Current — actively developed |
| 📦 Amazon Nova | Previous-gen multimodal understanding models | ✅ Available |
| 🖼️ Nova Canvas | Image generation | ✅ Available |
| 🎬 Nova Reel | Video generation | ✅ Available |
| 📜 Amazon Titan | First-gen text, embedding, and multimodal models | ⚠️ Legacy — no longer primary focus |
📊 Nova Model Overview
Section titled “📊 Nova Model Overview”| 🤖 Model | 🗂️ Category | 📄 Context | 🎯 Best For | ⚡ Speed |
|---|---|---|---|---|
| 🔬 Nova 2 Pro | Reasoning + multimodal | 1M tokens | Hard reasoning, complex analysis | Medium |
| ⚡ Nova 2 Lite | Reasoning + fast | 1M tokens | Everyday reasoning tasks | Fast |
| 🔊 Nova 2 Sonic | Speech-to-speech | 1M tokens | Real-time voice conversations | Real-time |
| 💎 Nova Premier | Multimodal (advanced) | 300K tokens | Complex reasoning + model distillation | Medium |
| ⚖️ Nova Pro | Multimodal | 300K tokens | Accuracy + speed + agentic workflows | Medium |
| ⚡ Nova Lite | Multimodal (fast) | 300K tokens | Real-time interactions, cost-efficient | Fast |
| ⚡ Nova Micro | Text-only | 128K tokens | Lowest latency, lowest cost | Fastest |
🚀 Amazon Nova 2 — Current Reasoning Generation
Section titled “🚀 Amazon Nova 2 — Current Reasoning Generation”The Nova 2 series introduced 1M token context, built-in tools (code interpreter, web grounding, MCP), and native reasoning capability.
🔬 Nova 2 Pro — Highest Intelligence (Nova 2 Series)
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Reasoning + multimodal — highest Nova 2 intelligence |
| 📄 Context window | 1,000,000 tokens |
| 👁️ Inputs | Text · images · video · speech — all in one reasoning flow |
| 🔧 Built-in tools | Code interpreter · web grounding · remote MCP tool support |
| 🌐 Access | Amazon Bedrock (US East, US West, Asia Pacific Tokyo) |
✅ Ideal for:
- 🧠 Complex multi-step reasoning across text, images, and video
- 🏗️ Agentic workflows requiring tool use and long context
- 📄 Large document analysis (contracts, reports, codebases)
- 🔬 Research and technical analysis tasks
- 💡 Training and distilling smaller custom models (teacher model role)
💡 Example:
“Analyse this 90-minute product demo video and this 500-page technical specification. Identify the three biggest gaps between what was demonstrated and what the spec requires.”
⚡ Nova 2 Lite — Everyday Reasoning (Fast)
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Reasoning — fast and cost-effective |
| 📄 Context window | 1,000,000 tokens |
| 🔧 Built-in tools | Code interpreter · web grounding · MCP support |
| 🌐 Access | Amazon Bedrock · EU regions available |
✅ Ideal for:
- 💬 Production chat assistants with reasoning capability
- 💻 Code generation and explanation
- 📝 Summarisation and document drafting
- 🏷️ Structured extraction with reasoning
- ⚡ Most everyday AI tasks where Nova 2 Pro is more than needed
🔊 Nova 2 Sonic — Speech-to-Speech Conversations
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Speech-to-speech — real-time conversational AI |
| 📄 Context window | 1,000,000 tokens |
| 🌍 Languages | 7 languages |
| 🌐 Access | Amazon Bedrock · EU regions available |
🌟 Key capabilities:
- 🎙️ Unifies speech understanding and speech generation in a single model
- 🔀 Cross-modal interaction — users can switch between voice and text mid-session
- 🗣️ Polyglot voices — adapts voice characteristics across languages
- 🔧 Asynchronous tool use — can call APIs while speaking
✅ Ideal for:
- 🤖 Real-time voice agents and conversational bots
- 📞 Customer service automation with natural speech
- 🌍 Multilingual voice assistants
- ♿ Accessibility and voice-first interfaces
💡 Three-piece voice → Nova 2 Sonic all-in-one:
Older voice systems required separate speech-to-text, LLM, and text-to-speech models. Nova 2 Sonic handles the entire spoken conversation in a single model.
📦 Amazon Nova — Previous Generation (Still Available)
Section titled “📦 Amazon Nova — Previous Generation (Still Available)”💎 Nova Premier — Most Capable Multimodal
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Multimodal — maximum capability |
| 📄 Context window | 300,000 tokens |
| 🎯 Special role | Best teacher model for distilling custom Nova variants |
| 👁️ Inputs | Text · images · video |
| 🌐 Access | Amazon Bedrock |
✅ Ideal for:
- 🏗️ Complex reasoning tasks requiring the full previous-gen Nova capability
- 🔧 Distilling customised Nova models via Nova Forge
- 🤖 High-stakes agentic workflows
⚖️ Nova Pro — Accuracy + Speed Balance
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Multimodal — best accuracy/speed combination |
| 📄 Context window | 300,000 tokens |
| 👁️ Inputs | Text · images · video |
| 🌐 Access | Amazon Bedrock |
✅ Ideal for:
- 💬 Production assistants needing multimodal understanding
- 🤖 Agentic workflows with API and tool calls
- 📊 Document and visual data analysis
⚡ Nova Lite — Fast Multimodal
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Multimodal — lightning fast, very low cost |
| 📄 Context window | 300,000 tokens |
| 👁️ Inputs | Text · images (multiple) · video (up to 30 min) |
| 🌐 Access | Amazon Bedrock |
✅ Ideal for:
- ⚡ Real-time customer interactions
- 👁️ Document and visual Q&A at scale
- 🏷️ Multimodal classification and extraction
⚡ Nova Micro — Fastest, Text-Only
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Text-only — lowest latency, lowest cost |
| 📄 Context window | 128,000 tokens |
| 👁️ Inputs | Text only |
| 🌐 Access | Amazon Bedrock |
✅ Ideal for:
- ⚡ Ultra-low-latency text tasks
- 🏷️ Classification, routing, summarisation, translation
- 💬 Interactive chat where speed beats all else
- 💰 Maximum cost efficiency on pure text workloads
🖼️ Nova Canvas — Image Generation
Section titled “🖼️ Nova Canvas — Image Generation”🖼️ Nova Canvas — Text-to-Image
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Image generation |
| 🎯 Purpose | Studio-quality images from text prompts |
| 🌐 Access | Amazon Bedrock |
✅ Ideal for:
- 🖼️ Marketing campaign visuals
- 🛍️ Product visualisation and photography
- 🎨 Creative content at scale
- 📱 Social media visuals
🎬 Nova Reel — Video Generation
Section titled “🎬 Nova Reel — Video Generation”🎬 Nova Reel — Text-to-Video
| 🔧 Property | 📋 Detail |
|---|---|
| 🏷️ Category | Video generation |
| 🎯 Purpose | Generate video clips from text prompts |
| 🌐 Access | Amazon Bedrock |
✅ Ideal for:
- 🎬 Short-form marketing video
- 📱 Social media content
- 🎨 Creative concept visualisation
📜 Amazon Titan — Legacy Models
Section titled “📜 Amazon Titan — Legacy Models”The Titan series was Amazon’s first-generation model family. It is no longer actively developed but remains available on Bedrock for existing workloads.
📜 Amazon Titan Text Models (Legacy)
| 🤖 Model | 📄 Context | 🎯 Focus |
|---|---|---|
| Titan Text Premier | 32K tokens | General text, complex workflows, fine-tuning |
| Titan Text Express | 8K tokens | 100+ languages, general-purpose |
| Titan Text Lite | 4K tokens | Lightweight summarisation and simplification |
⚠️ Titan is legacy — for new projects, use Nova 2 Lite or Nova Micro instead.
🔍 Amazon Titan Embedding Models (Legacy)
| 🤖 Model | 📐 Dimensions | 🎯 Focus |
|---|---|---|
| Titan Embeddings G1 | Up to 1,024 | Text-to-vector for semantic search and RAG |
| Titan Embeddings V2 | Up to 1,024 | Updated text embeddings |
| Titan Multimodal Embeddings | 256 / 384 / 1,024 | Unified text + image vectors |
✅ Still useful for:
- 🔍 Semantic search on existing pipelines built with Titan embeddings
- 🔄 Cross-modal image + text recommendation systems
⚠️ For new embedding work, also evaluate OpenAI or Cohere embedding models available on Bedrock.
🔧 Nova Forge — Custom Model Training
Section titled “🔧 Nova Forge — Custom Model Training”Nova Forge allows organisations to train custom variants of Nova models using their own proprietary data.
| ✅ Capability | 📋 Detail |
|---|---|
| 🔧 Access | Pre-trained, mid-trained, and post-trained Nova checkpoints |
| 📊 Data blending | Mix proprietary data with Amazon Nova training datasets |
| 🎯 Purpose | Build domain-specialised Nova models (legal, medical, finance) |
| 🌐 Access | Amazon Bedrock |
🎯 Which Amazon Model to Use
Section titled “🎯 Which Amazon Model to Use”| 🎯 Task | ✅ Best Model | 💡 Why |
|---|---|---|
| 🧠 Complex reasoning over text + video + images | Nova 2 Pro | Highest intelligence + 1M context + all modalities |
| 💬 Everyday production assistant | Nova 2 Lite | Reasoning + 1M context at lower cost |
| 🔊 Real-time voice conversation | Nova 2 Sonic | All-in-one speech-to-speech model |
| 🤖 Agentic workflows with tools | Nova 2 Pro or Nova 2 Lite | Built-in code interpreter, web grounding, MCP |
| ⚡ Ultra-fast text classification or routing | Nova Micro | Cheapest and fastest text model |
| 📄 Long document analysis (up to 300K) | Nova Pro or Nova Lite | Strong multimodal at 300K context |
| 🖼️ Generate images for marketing | Nova Canvas | Amazon’s native image generation |
| 🎬 Generate short video clips | Nova Reel | Amazon’s native video generation |
| 🔧 Build a domain-specific Nova model | Nova Forge + Nova Premier | Custom training on proprietary data |
| 📜 Maintain existing Titan pipeline | Titan (appropriate tier) | Backwards compatibility |
🌐 Where to Access Amazon AI Models
Section titled “🌐 Where to Access Amazon AI Models”| 🏷️ Platform | 🤖 Models Available |
|---|---|
| 🟠 Amazon Bedrock | All Nova and Titan models |
| 🔷 AWS Console | Amazon Bedrock management UI |
| 🔌 AWS SDK | Programmatic API access to all models |
| 🌍 Regions | US East (N. Virginia) · US West (Oregon) · Asia Pacific (Tokyo) · EU (Nova 2) |