Amazon Titan was Amazon’s first own model family on AWS Bedrock. Its successor is the Amazon Nova family, whose Nova 2 generation has been available on Bedrock since December 2025: Nova 2 Lite, Nova 2 Sonic and, since October 2026, Nova 2.5 Sonic are generally available, while Nova 2 Pro and Nova 2 Omni run as previews. Titan is now relevant only for embeddings: the Titan text-generation models are no longer listed on Bedrock, and the image model Titan Image Generator v2 reached end of life on 30 June 2026.
innFactory AI Consulting from Rosenheim advises companies in the DACH region on GDPR-compliant use of Amazon Nova and Titan via AWS Bedrock, including in Frankfurt. This overview reflects the state as of October 2026.
Current models at a glance
| Model | Status | Focus | EU availability per AWS model card |
|---|---|---|---|
| Nova 2 Lite | GA | Reasoning, documents, 1M context | EU geo cross-region profile eu.amazon.nova-2-lite-v1:0 |
| Nova 2.5 Sonic | GA | Real-time speech-to-speech with tool calling | In-Region Stockholm (eu-north-1) |
| Nova 2 Sonic | GA | Real-time speech-to-speech (predecessor) | In-Region Stockholm (eu-north-1) |
| Nova 2 Pro | Preview (early access) | Complex workloads, agentic workflows | – |
| Nova 2 Omni | Preview (early access) | Multimodal reasoning | – |
| Nova Multimodal Embeddings | GA | Embeddings for text, image, video, audio | none (us-east-1 only) |
| Titan Text Embeddings V2 | Active | Text embeddings | In-Region Frankfurt, Stockholm, Milan, Spain, Ireland, Paris (plus Zurich and London, UK) |
| Titan Embeddings G1 - Text | Active | Text embeddings (predecessor) | In-Region Frankfurt |
| Titan Multimodal Embeddings G1 | Active | Text-image embeddings | In-Region Frankfurt, Ireland, Paris (plus London, UK) |
Nova 2 Lite
- 1M token context window (Titan: 32K)
- Built-in tools: code interpreter, web grounding, MCP support
- Fast, cost-efficient reasoning tasks
- Usable in the EU via a dedicated EU geo cross-region profile that routes requests exclusively between Frankfurt, Stockholm, Milan, Spain, Ireland and Paris (data stays in the EU, but with no guarantee of a single region)
Nova 2.5 Sonic and Nova 2 Sonic
Nova 2.5 Sonic has been Amazon’s current speech-to-speech model since 5 October 2026. Per AWS, it improves reasoning, instruction following and tool-calling accuracy over Nova 2 Sonic, responds with lower latency and calls tools asynchronously without interrupting the conversation. Text and voice can be mixed in one session, and the context window covers 256,000 tokens. Per the model card, Nova 2.5 Sonic does not support Guardrails or Knowledge Bases.
Both Sonic models run in-Region in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo) and Stockholm (eu-north-1), the only EU region with genuine data residency for these models. Nova 2 Sonic (December 2025) remains active per AWS.
Nova 2 Pro and Nova 2 Omni (preview)
Nova 2 Pro is designed per Amazon for complex workloads (multi-step reasoning, agentic workflows, video reasoning), Nova 2 Omni for multimodal reasoning across text, image, video and speech, and generates text and images. Both are currently available via early access only.
Nova Act and Nova Forge
Nova Act automates browser-based UI workflows, Nova Forge is for building your own frontier models on a Nova base.
Titan: embeddings only
Per AWS’s model overview, the Titan embedding models (Titan Text Embeddings V2, Titan Text Embeddings G1) and Titan Multimodal Embeddings G1 remain active, each with lifecycle status “Active”.
| Model | Dimensions | Feature |
|---|---|---|
| Titan Text Embeddings V2 | 256/512/1024 | Multilingual, flexible dimensions |
| Titan Multimodal Embeddings G1 | 1024/384 | Text + image embeddings |
EU availability and GDPR
- EU data residency for Titan embeddings: per the AWS model card they run in-Region, including Frankfurt (eu-central-1); data stays in the EU.
- Nova 2 Lite: via the EU geo cross-region profile – data stays in the EU, but with no commitment to a single region. Nova 2.5 Sonic and Nova 2 Sonic: in-Region in Stockholm.
- Nova Multimodal Embeddings: per the AWS model card us-east-1 only, no EU option.
- DPA: a data processing agreement with AWS is available.
- AWS integration: IAM, VPC endpoints, CloudWatch, guardrails and fine-tuning on your own data; SOC and ISO certifications and HIPAA eligibility per AWS.
Model lifecycle
- Titan Text Premier, Express and Lite (context 32K, 8K and 4K) are no longer listed as available Amazon models in Bedrock’s current model overview, and the model cards return no content. We read this as an effective retirement on Bedrock; we found no officially documented end date.
- Titan Image Generator v2 (up to 2048x2048) had an EOL date of 30 June 2026 per the AWS model card; the model card has since been removed. The model was never available in an EU region (only us-east-1, us-west-2).
- Nova Canvas and Nova Reel (image and video generation, December 2024) reached end of life on 30 September 2026 per the AWS lifecycle page and were removed from all regions (previously us-east-1, Ireland and Tokyo). Bedrock therefore no longer offers a generally available Amazon model for image or video generation; Nova 2 Omni (preview) generates images as well as text.
- Nova Sonic (first generation) reached end of life on 14 September 2026; its successors are Nova 2 Sonic and Nova 2.5 Sonic.
- Nova Premier (not part of Nova 2) reached end of life on Bedrock on 14 September 2026 per the AWS model card and was removed from all regions; migrate to Nova 2 Pro or Nova Pro.
Use cases
- Enterprise RAG: Titan Text Embeddings V2 for semantic search with in-Region operation in the EU, Nova Multimodal Embeddings for text, images, video and audio (us-east-1 only); Bedrock Knowledge Bases and OpenSearch for vector search.
- Document processing: Nova 2 Lite with 1M context for very long documents, summarisation and structured extraction.
- Content generation: marketing copy, product descriptions, translations.
- Voice applications: Nova 2.5 Sonic for customer service and voice agents.
Integration with CompanyGPT
Amazon Nova and Titan are integrated into CompanyGPT via AWS Bedrock – ideal for companies with existing AWS infrastructure that need a GDPR-compliant solution.
Our recommendation
Amazon Nova 2 Lite is our default recommendation for AWS-native enterprises: 1M context, built-in tools and EU availability via a dedicated EU geo profile. For embeddings with EU data residency we recommend Titan Text Embeddings V2 with genuine in-Region hosting, including in Frankfurt; Nova Multimodal Embeddings also handles images, video and audio, but per the AWS model card runs only in us-east-1. For voice applications we recommend Nova 2.5 Sonic with in-Region operation in Stockholm. For new text-generation projects there is no way around Nova, since the Titan text models are no longer available.
For demanding text tasks we recommend Claude on AWS Bedrock, for self-hosting Llama via SageMaker. We support you in selecting and integrating the right Bedrock models.