EU Region Model Support and Mapping

Overview

The GenAI-IDP accelerator supports deployment in EU regions with automatic model mapping between US and EU model endpoints. This document outlines the available models, mapping behavior, and fallback mechanisms.

Complete Model Mappings

The following table shows all US to EU model mappings currently configured in the system:

US Model	EU Model	Notes
`us.amazon.nova-lite-v1:0`	`eu.amazon.nova-lite-v1:0`	Direct mapping
`us.amazon.nova-pro-v1:0`	`eu.amazon.nova-pro-v1:0`	Direct mapping
`us.amazon.nova-premier-v1:0`	`eu.anthropic.claude-sonnet-4-5-20250929-v1:0`	Fallback mapping
`us.amazon.nova-2-lite-v1:0`	`eu.amazon.nova-2-lite-v1:0`	Direct mapping
`us.anthropic.claude-3-haiku-20240307-v1:0`	`eu.anthropic.claude-3-haiku-20240307-v1:0`	Direct mapping
`us.anthropic.claude-3-5-haiku-20241022-v1:0`	`eu.anthropic.claude-sonnet-4-5-20250929-v1:0`	Fallback mapping
`us.anthropic.claude-haiku-4-5-20251001-v1:0`	`eu.anthropic.claude-haiku-4-5-20251001-v1:0`	Direct mapping
`us.anthropic.claude-3-5-sonnet-20241022-v2:0`	`eu.anthropic.claude-3-5-sonnet-20241022-v2:0`	Direct mapping
`us.anthropic.claude-3-7-sonnet-20250219-v1:0`	`eu.anthropic.claude-3-7-sonnet-20250219-v1:0`	Direct mapping
`us.anthropic.claude-sonnet-4-20250514-v1:0`	`eu.anthropic.claude-sonnet-4-20250514-v1:0`	Direct mapping
`us.anthropic.claude-sonnet-4-5-20250929-v1:0`	`eu.anthropic.claude-sonnet-4-5-20250929-v1:0`	Direct mapping
`us.anthropic.claude-sonnet-4-6`	`eu.anthropic.claude-sonnet-4-6`	Direct mapping
`us.anthropic.claude-sonnet-4-6:1m`	`eu.anthropic.claude-sonnet-4-6:1m`	Direct mapping
`us.anthropic.claude-opus-4-20250514-v1:0`	`eu.anthropic.claude-sonnet-4-5-20250929-v1:0`	Fallback mapping
`us.anthropic.claude-opus-4-1-20250805-v1:0`	`eu.anthropic.claude-sonnet-4-5-20250929-v1:0`	Fallback mapping
`us.anthropic.claude-opus-4-5-20251101-v1:0`	`eu.anthropic.claude-opus-4-5-20251101-v1:0`	Direct mapping
`us.anthropic.claude-opus-4-6-v1`	`eu.anthropic.claude-opus-4-6-v1`	Direct mapping
`us.anthropic.claude-opus-4-6-v1:1m`	`eu.anthropic.claude-opus-4-6-v1:1m`	Direct mapping
`us.anthropic.claude-opus-4-7`	`eu.anthropic.claude-opus-4-7`	Direct mapping
`us.anthropic.claude-opus-4-7:1m`	`eu.anthropic.claude-opus-4-7:1m`	Direct mapping
`us.anthropic.claude-opus-4-8`	`eu.anthropic.claude-opus-4-8`	Direct mapping
`us.anthropic.claude-opus-4-8:1m`	`eu.anthropic.claude-opus-4-8:1m`	Direct mapping
`us.meta.llama4-maverick-17b-instruct-v1:0`	`eu.anthropic.claude-sonnet-4-5-20250929-v1:0`	Fallback mapping
`us.meta.llama4-scout-17b-instruct-v1:0`	`eu.anthropic.claude-sonnet-4-5-20250929-v1:0`	Fallback mapping

Mapping Types

Direct Mapping: US model has exact EU equivalent
Fallback Mapping: US model not available in EU, mapped to best available EU alternative

US-Only Models (not offered in EU deployments)

The OpenAI frontier models are served via the bedrock-mantle endpoint and are only available in US (and us-gov) regions — there is no EU availability and no cross-region/geo inference. Rather than mapping them to a Claude fallback, the UpdateConfiguration Lambda removes them from the model dropdowns when the stack is deployed in an EU region, so they are never offered where they cannot be called.

Model	Availability
`openai.gpt-5.4`	`us-east-2`, `us-west-2`, `us-gov-west-1`
`openai.gpt-5.5`	`us-east-2`

See OpenAI GPT-5.x model support below.

Available EU Models

Based on the mappings above, the following EU models are supported:

Amazon Nova Models

eu.amazon.nova-lite-v1:0
eu.amazon.nova-pro-v1:0

Anthropic Claude Models

eu.anthropic.claude-3-haiku-20240307-v1:0
eu.anthropic.claude-haiku-4-5-20251001-v1:0
eu.anthropic.claude-3-5-sonnet-20241022-v2:0
eu.anthropic.claude-3-7-sonnet-20250219-v1:0
eu.anthropic.claude-sonnet-4-20250514-v1:0
eu.anthropic.claude-sonnet-4-5-20250929-v1:0

Model Mapping Behavior

Automatic Region Detection

The system automatically detects the deployment region and applies appropriate model mappings:

EU Regions: Any region starting with eu- (e.g., eu-west-1, eu-central-1)
US Regions: Any region starting with us- (e.g., us-east-1, us-west-2)

Mapping Logic

US to EU Mapping: When deployed in EU regions, US model IDs are automatically mapped to their EU equivalents
EU to US Mapping: When deployed in US regions, EU model IDs are mapped back to US equivalents
Fallback Behavior: If no mapping exists, the original model ID is returned unchanged

Configuration Processing

The UpdateConfiguration lambda processes default configurations as follows:

Model ID Detection: Identifies strings containing us. or eu. prefixes
Region-Based Swapping: Swaps model IDs based on deployment region
Logging: Logs all model swaps for debugging purposes

Important Limitations

Fallback Mappings

⚠️ Critical: Some US models are not directly available in EU regions and use fallback mappings:

Nova Premier: us.amazon.nova-premier-v1:0 → eu.anthropic.claude-sonnet-4-5-20250929-v1:0
Claude 3.5 Haiku: us.anthropic.claude-3-5-haiku-20241022-v1:0 → eu.anthropic.claude-sonnet-4-5-20250929-v1:0
Claude Opus Models: Both Opus variants → eu.anthropic.claude-sonnet-4-5-20250929-v1:0

Implications of Fallback Mappings

Performance Changes: Fallback models may have different speed/cost characteristics
Behavior Differences: Different models may produce varying outputs for the same input
Cost Impact: Fallback models may have different pricing structures

Third-Party Models (US-Only)

The following third-party models are available in US regions only and have no EU cross-region inference profiles. In EU deployments, the Meta Llama 4 models fall back to Claude Sonnet 4.5, while the single-region models (Gemma, Nemotron) are not available and should not be selected:

Meta Llama 4 Maverick 17B: us.meta.llama4-maverick-17b-instruct-v1:0 → eu.anthropic.claude-sonnet-4-5-20250929-v1:0 (fallback)
Meta Llama 4 Scout 17B: us.meta.llama4-scout-17b-instruct-v1:0 → eu.anthropic.claude-sonnet-4-5-20250929-v1:0 (fallback)
Google Gemma 3 27B IT: google.gemma-3-27b-it — single-region only, no EU availability
NVIDIA Nemotron Nano 12B v2: nvidia.nemotron-nano-12b-v2 — single-region only, no EU availability
Qwen3 VL 235B A22B: qwen.qwen3-vl-235b-a22b — single-region only, no EU availability

Missing Direct EU Equivalents

The following US models do not have direct EU equivalents:

Nova Premier
Claude 3.5 Haiku (specific version)
Claude Opus 4 variants
Meta Llama 4 Maverick/Scout (US cross-region only)
Google Gemma 3, NVIDIA Nemotron, Qwen (single-region US only)

Best Practices

For EU Deployments

Verify Model Availability: Ensure all configured models have EU mappings before deployment
Test Thoroughly: Test document processing with actual EU model endpoints
Monitor Logs: Check UpdateConfiguration lambda logs for model swap confirmations

Recommended EU Models

Fastest: eu.amazon.nova-lite-v1:0
Balanced: eu.amazon.nova-pro-v1:0
Best Quality: eu.anthropic.claude-sonnet-4-5-20250929-v1:0

Troubleshooting

Common Issues

Model Not Found Errors: Check if the model has an EU mapping
Access Denied: Verify Bedrock model access is enabled for EU models
Unexpected US Model Usage: Check logs for missing mappings

Region Availability

This feature supports all AWS regions but is specifically designed for:

EU Regions: eu-west-1, eu-central-1, eu-north-1, etc.
US Regions: us-east-1, us-west-2, etc.

Other regions (Asia-Pacific, etc.) will use the original model IDs without mapping.

OpenAI GPT-5.x Models (bedrock-mantle)

See OpenAI GPT-5.x Models for the authoritative, full support matrix and caveats. This section summarizes how they differ and why they are US-only (and therefore hidden in EU-region deployments).

OpenAI’s frontier models (openai.gpt-5.4, openai.gpt-5.5) are integrated differently from every other supported model.

How they differ

Aspect	Claude / Nova (Converse)	OpenAI GPT-5.x (Responses)
API	`bedrock-runtime` Converse	`bedrock-mantle` OpenAI Responses API (REST)
Endpoint	boto3 `converse()`	`https://bedrock-mantle.{region}.api.aws/openai/v1/responses`
Auth	IAM (boto3)	IAM via SigV4-signed HTTP (signing name `bedrock-mantle`)
Inference params	temperature / top_p / top_k	reasoning models — sampling params omitted; `reasoning.effort`
Prompt caching	cachePoint supported	not supported (`<<CACHEPOINT>>` stripped)
Agentic extraction	supported	not supported (rejected by config-validate; raises at runtime)
Service tiers	priority / flex / standard	standard only
Region prefix / `:1m`	yes	none

The accelerator’s idp_common.bedrock.BedrockClient.invoke_model detects these model IDs and routes them to idp_common/bedrock/openai_responses.py, which translates the same inputs to/from the Converse-shaped contract — so all services (OCR, classification, extraction, assessment, summarization, evaluation) use them with no special handling. Chat-with-Document also supports them: the chat processor streams GPT-5.x responses via the Responses API’s SSE stream (stream_responses_api in openai_responses.py), publishing incremental token deltas to the UI exactly like the Converse streaming path.

Regional availability

Model	In-Region availability	Context window
`openai.gpt-5.5`	`us-east-2`	272K
`openai.gpt-5.4`	`us-east-2`, `us-west-2`, `us-gov-west-1`	272K

There is no EU availability and no geo/global cross-region inference. When the IDP stack is deployed in a region where the model is not available, the request is routed to a known-available bedrock-mantle region (a warning is logged about cross-region data movement). EU-region deployments hide these models from the dropdowns entirely.

Reasoning effort

GPT-5.x are reasoning models — they reject temperature/top_p/top_k and are instead tuned via reasoning effort. Each model-selectable service (OCR, classification, extraction, assessment, summarization, evaluation) exposes a reasoning_effort config field with allowed values minimal, low, medium, high (default medium). It is an OpenAI-only parameter — ignored by Anthropic/Nova and other model families. Set it per service in the config (or the UI editor); for example, in extraction:

extraction:
  model: "openai.gpt-5.4"
  reasoning_effort: "high"

Configuration / environment variables

Variable	Purpose	Default
`BEDROCK_MANTLE_REGION`	Pin the bedrock-mantle region for all GPT-5.x calls	derived from stack region with per-model fallback
`BEDROCK_MANTLE_SIGNING_NAME`	SigV4 signing service name	`bedrock-mantle`
`BEDROCK_MANTLE_REASONING_EFFORT`	Global fallback reasoning effort when a service config omits `reasoning_effort`	`medium`

IAM

Lambda execution roles that perform generation are granted bedrock-mantle:CreateInference (plus GetProject / ListProjects / ListTagsForResources), equivalent to the AWS-managed AmazonBedrockMantleInferenceAccess policy. When using cross-account Bedrock via a hub-account role, that role must also grant these bedrock-mantle actions (see Cross-Account Bedrock).