Amazon Bedrock Model Profiler: open-source model catalog

Amazon Bedrock Model Profiler is an open source web application that aggregates model metadata from multiple AWS APIs and public sources into a single, searchable interface for selecting foundation models on Amazon Bedrock. The profiler consolidates specifications, pricing, quotas and regional availability so teams can compare options without manual cross-referencing.

How does the Model Profiler collect and present model data?

The Model Profiler collects data from seven distinct sources: five AWS APIs and two public URLs, processes it through a serverless pipeline, and publishes two production JSON files served from Amazon S3 and CloudFront. The pipeline is orchestrated by AWS Step Functions, uses 17 AWS Lambda functions across four phases, and writes bedrock_models.json and bedrock_pricing.json for the frontend.

The seven sources are the Amazon Bedrock ListFoundationModels API, AWS Price List API, AWS Service Quotas API, Amazon Bedrock ListInferenceProfiles API, Amazon Bedrock Mantle API, the LiteLLM Model Database (public URL), and AWS documentation (public URL). The pipeline runs daily at 6 AM UTC, completes in 8–12 minutes, and reduces API calls from approximately 480 to 29 per execution by using inter-Lambda S3 caching, representing a 97% cache hit rate.

What features does the Model Profiler provide for choosing models?

Model Profiler offers a searchable Model Explorer, side-by-side Model Comparison for up to 25 models, a Regional Availability matrix across 33 Bedrock-enabled regions, and a Favorites shortlist for tracking candidates. The explorer shows technical specifications, pricing breakdowns by Region and consumption type, service quotas (TPM and RPM), and lifecycle status for each model.

The Model Explorer lists 120+ foundation models and filters by provider (Anthropic, Meta, Amazon, Mistral AI, Cohere, OpenAI and others), capabilities such as vision, code generation, function calling, or embeddings, and input/output modalities. Consumption options include In Region on-demand inference, Cross-Region Inference Service (CRIS), Batch processing, and Mantle managed inference. Comparison view includes per-token pricing, batch discounts (ranging from 30-50% depending on model and modality), context window sizes, maximum output tokens, and capability matrices.

Why it matters

Model selection on a multi-provider platform is fragmented across console pages, docs, and regional API calls; the Profiler centralizes that information and removes much of the manual work. For teams evaluating new workloads, optimizing cost and throughput, or planning multi-region deployments, the consolidated catalog, daily refresh cadence and the Regional Availability matrix replace dozens of individual API calls and manual cross-referencing.

The pipeline’s operational metrics are concrete signals of scale and efficiency: it gathers regional availability across 33 regions, runs every day at 6 AM UTC, and completes in 8–12 minutes while achieving a 97% cache hit rate that trims API calls from about 480 to 29 per run.

What to watch

Check for updates to the published JSON files bedrock_models.json and bedrock_pricing.json to see new models, pricing changes, or lifecycle status updates after the daily run. Also watch whether the profiler’s regional coverage changes as Amazon Bedrock expands into additional regions, since the pipeline dynamically discovers regions rather than using a hardcoded list.