Amazon Bedrock's open-source Model Profiler simplifies AI model selection

Amazon Bedrock, AWS's managed service for building production-ready AI applications, now offers the open-source Model Profiler to tackle the complexity of choosing among over 100 foundation models from Anthropic, OpenAI, Meta, Mistral AI, Cohere, and Amazon. Instead of scouring multiple console pages and documentation for capabilities, pricing, regional availability, context window limits, and throughput quotas, teams get a single, searchable interface. The tool aggregates data from five AWS APIs (including ListFoundationModels, Price List, Service Quotas, ListInferenceProfiles, and Mantle API) plus two public URLs (LiteLLM Model Database and AWS Documentation). Behind the scenes, a fully automated serverless pipeline orchestrated by AWS Step Functions runs daily at 6 AM UTC, completing in 8–12 minutes. Seventeen AWS Lambda functions process data across four phases, using inter-Lambda S3 caching to slash API calls from roughly 480 to 29 (a 97% cache hit rate). A self-healing agentic system powered by Amazon Bedrock detects data gaps and applies safe configuration fixes automatically.

The Model Profiler enables advanced filtering and side-by-side comparisons with detailed model cards that include token-per-minute (TPM) and requests-per-minute (RPM) quotas—critical metrics for throughput ceilings and API call limits. Regional availability maps show which models are supported across 33 regions, and pricing is updated daily. Users can deploy the entire tool in under five minutes using the provided AWS CloudFormation template, which includes an Amazon EventBridge rule (configurable for frequency). Key real-world scenarios include evaluating models for new workloads, optimizing cost and performance, and migrating from other AI systems. The pipeline dynamically discovers new AWS regions as they launch, so no hardcoded lists are needed. Whether you're comparing context window sizes, inference configurations for cross-region setups, or model lifecycle statuses (active, legacy, end-of-life), Model Profiler streamlines decision-making that previously required manual effort across dispersed sources.