Models
The gateway routes across the leading Amazon Bedrock models. Use auto, auto:cost or
auto:quality to let it choose, or pin one by its id.
| Id | Provider | Context / max output | Capabilities |
|---|---|---|---|
amazon.nova-micro | Amazon | 128K / 10K | JSON, tools |
amazon.nova-lite | Amazon | 300K / 10K | JSON, tools, vision |
amazon.nova-pro | Amazon | 300K / 10K | JSON, tools, vision |
anthropic.claude-haiku-5-5 | Anthropic | 200K / 64K | JSON, tools, vision |
anthropic.claude-sonnet-5-5 | Anthropic | 200K / 64K | JSON, tools, vision |
anthropic.claude-opus-5-5 | Anthropic | 200K / 64K | JSON, tools, vision |
meta.llama3-3-70b | Meta | 128K / 8K | JSON, tools |
openai.gpt-oss-120b | OpenAI | 128K / 32K | JSON, tools |
Catalog version: bedrock-2026-10-18-r2. Each response's x-acg-catalog-version header names
the catalog used for its choice.
Which ones you can use
The gateway picks only from models enabled in your AWS account, in the region you connected.
GET /models lists the routing modes and every model id available to you:
curl https://api.aicloudgateway.com/v1/models -H "Authorization: Bearer $ACG_API_KEY"
Amazon Nova, Meta Llama and OpenAI gpt-oss work on first use. Claude models need a one-time approval in the Bedrock console; see Your AWS account.
Regions: us-east-1, us-east-2 and us-west-2. openai.gpt-oss-120b is available in
us-east-1 and us-west-2.
Tokens are billed by AWS
Model usage is billed to your AWS account at your Bedrock prices. AI Cloud Gateway adds no markup. See Plans and billing.
Full guide: routing.md on GitHub