Model Catalog
The model catalog is a platform-global DynamoDB entity maintained by copebit ops. It maps Anthropic model IDs to Bedrock inference profile ARNs per region, with pricing and availability metadata.
Storage
All catalog entries share PK=MODELS, so the entire catalog is fetched
with one Query (PK=MODELS, SK begins_with MODEL#). The sort key is
MODEL#<model_id> (e.g. MODEL#claude-opus-4-7).
Fields
| Field | Notes |
|---|---|
model_id | Anthropic model ID, e.g. claude-opus-4-7 |
family | opus / sonnet / haiku |
display_name | Human-readable, e.g. “Claude Opus 4.7” |
version | Semver, e.g. 4.7 |
bedrock_inference_profile_arns | Region → ARN map |
available_regions | Regions where the model is available in Bedrock |
status | ga / preview / deprecated |
pricing | { input_per_1k, output_per_1k, cache_read_per_1k, cache_creation_per_1k } USD |
is_latest | Whether this is the *-latest for its family |
*-latest aliases
Grants may list claude-opus-latest, claude-sonnet-latest, or
claude-haiku-latest. The proxy resolves these at request time to the
newest status=ga version in that family. Resolution is cached
in-memory for 5 minutes.
Exact model IDs, versions, and Bedrock ARNs are verified against live
Bedrock availability at launch. The catalog is re-seeded by copebit ops
as AWS adds versions (scripts/seed_models.py).