Model Info for Amazon Bedrock

A lightweight Python helper to simplify retrieving metadata about AI models on Amazon Bedrock.

This early release is focussed on fetching pricing information in particular, although we hope to expand to other metadata in future and contributions are welcome!

⚠️ WARNING: This helper library is provided "as is", as described in the LICENSE.

• The data aggregations it performs are non-trivial and may contain bugs.
• No warranty is provided or liability accepted for the correctness of results.
• Users should validate against authoritative sources like the Amazon Bedrock Pricing page and Amazon Bedrock model cards before using results for contractual, billing-critical, or other high-impact decisions.

Although the underlying AWS Price List API provides authoritative information, its listings for Amazon Bedrock are distributed across multiple service codes, and their attributes don't correspond directly to unique model or inference profile IDs used by Amazon Bedrock inference APIs. This library provides a simpler way to look up current pricing records by model ID.

Installation

Because this library uses the AWS Price List API, the environment where it runs needs AWS credentials configured with appropriate IAM permissions to list services and products.

Install on Python 3.11+ with your preferred package manager. For example, with uv:

uv add model-info-amazon-bedrock

Or with pip:

pip install model-info-amazon-bedrock

Usage

Many factors can affect the final price of a foundation model on Amazon Bedrock, including for example:

• Which service tier you use for to balance between cost, speed, and predictability.
• Which Region or inference scope you target, including global cross-region inference which can be cheaper.
• Whether you use a longer-context variant of a model or the standard context length.
• Private pricing agreements or any other account-specific arrangements that are not represented by public list prices.

It's important to consider these complexities for detailed assessments and high-impact decisions.

However, this library mainly helps retrieve indicative list pricing, including standard on-demand pricing, by foundation model or cross-Region inference profile ID:

from model_info_amazon_bedrock import BedrockModelInfoClient

client = BedrockModelInfoClient()

pricing = client.get_model_pricing(
    "global.anthropic.claude-sonnet-5",
    region="us-east-1",
)

# Standard on-demand token pricing, when available.
print(f"Input:  ${pricing.input_tokens:.2f} per million tokens")
print(f"Output: ${pricing.output_tokens:.2f} per million tokens")

Example output:

Input:  $2.00 per million tokens
Output: $10.00 per million tokens

You can inspect all pricing dimensions matched and classified for the model:

for dim in pricing.dimensions:
    print(
        f"{dim.direction.value} ({dim.tier.value}, {dim.scope.value}): "
        f"${dim.price:.4f} per {dim.unit.value}"
    )

Price Records

get_model_pricing returns a ModelPricing object. Each PriceDimension in ModelPricing.dimensions contains:

The convenience properties input_tokens, output_tokens, input_images, and output_images return the standard tier pricing, or None when that value couldn't be identified. Use get_price or filter_dimensions when you need a specific tier, scope, cache operation, modality, or context length.

Matching and classification are heuristic. The returned dimensions reflect AWS Price List records the library recognized for the requested model/profile; they are not a guarantee that every AWS pricing dimension for every model is represented.

Caching

Pricing data is fetched lazily by source and cached per Region on each client instance:

client = BedrockModelInfoClient()

# First lookup fetches the required AWS pricing source data (so may take a few seconds)
pricing = client.get_model_pricing("anthropic.claude-sonnet-4-20250514-v1:0")

# Later lookups in the same Region reuse cached source data when possible (faster)
pricing = client.get_model_pricing("meta.llama3-70b-instruct-v1:0")

# Force fresh source data for this lookup.
pricing = client.get_model_pricing(
    "anthropic.claude-sonnet-4-20250514-v1:0",
    refresh=True,
)

# Clear cached pricing data for a Region.
client.invalidate_pricing_cache("us-east-1")

A failed refresh raises the underlying error and preserves existing cached data.

Custom boto3 Session

import boto3

from model_info_amazon_bedrock import BedrockModelInfoClient

session = boto3.Session(
    profile_name="my-profile",
    region_name="us-west-2",
)
client = BedrockModelInfoClient(session=session)

Error Handling

from model_info_amazon_bedrock import (
    BedrockModelInfoClient,
    PricingNotFoundError,
)

client = BedrockModelInfoClient()

try:
    pricing = client.get_model_pricing("nonexistent.model-id")
except PricingNotFoundError as exc:
    print(f"No pricing for {exc.model_id} in {exc.region}")
except ValueError as exc:
    print(f"Invalid input: {exc}")

AWS credential, service, throttling, and network exceptions from boto3/botocore propagate to the caller.

Development

See CONTRIBUTING.md for guidance including how best to engage with our community, setting up your local development environment for the project, tips on testing & debugging, and architectural guidance.

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.