Amazon Bedrock
100+ foundation models, one managed AWS API
- Overview
- Amazon Bedrock provides API access to 100+ AI models from 18+ providers — including Anthropic Claude, OpenAI GPT-5.6/GPT-6 Astra, Meta Llama 4, Grok 4.7, Amazon Nova, Mistral, and DeepSeek — through AWS-managed infrastructure with no servers to provision. Teams can fine-tune models on private data, build agentic pipelines via Bedrock AgentCore, and ground outputs with managed Knowledge Bases, all within AWS security and compliance boundaries. Pricing is token-based with five billing modes; a free tier is available for experimentation.
- Why it matters
- Bedrock has become the default enterprise AI infrastructure layer for AWS shops, displacing point solutions by consolidating model access, agent orchestration, RAG, and guardrails into one auditable platform. The April–September 2026 OpenAI partnership — adding GPT-5.6, GPT-6 Astra, and cross-region inferencing — means organizations no longer have to choose between frontier model access and AWS-native security controls. The key risk to watch is cost complexity: five billing modes and per-feature metering (Guardrails, Knowledge Base retrieval, AgentCore transitions) routinely inflate bills 20–30% beyond naive token estimates. Vendor lock-in to AgentCore's orchestration layer is a legitimate architectural concern for teams that may want to run multi-cloud in the future.
- In practice
- Typical enterprise teams use Bedrock as an internal LLM gateway: route cheap, fast queries to Amazon Nova Micro or Claude Haiku 4.5, escalate complex reasoning to a top-tier Claude Opus model or GPT-6 Astra, and run nightly document pipelines in batch mode at 50% off. RevOps and data engineering teams wire Knowledge Bases to S3-resident policy documents, then expose a RAG-backed assistant to internal users via Lambda. The honest limitation is observability: while CloudTrail logs API calls, per-token cost attribution across models and business units requires custom tagging and tooling that most teams underestimate at procurement time.
- Best for
- AWS-native enterprises that need a single control plane for multi-model AI applications with strict data governance, compliance, and agentic workflow requirements
- Pricing
- Free tier available for experimentation. On-demand pay-per-token from ~$0.035/M tokens (Amazon Nova Micro) up to $75/M tokens (top-tier frontier models). Batch inference at 50% off on-demand rates. Provisioned throughput and Enterprise (AWS EDP) plans available on request. Separate charges apply for Knowledge Base storage/retrieval, Guardrails, and AgentCore node transitions.
Key features
- 100+ foundation models from 18+ providers via a single API, including OpenAI, Anthropic, Meta, xAI, Mistral, and Amazon Nova
- AgentCore: managed agent harness, CLI, and web search grounding for production-grade agentic workflows without orchestration code
- Managed Knowledge Base with Smart Parsing and Agentic Retriever for enterprise RAG pipelines
- Guardrails for content filtering, PII redaction, and responsible AI controls
- Cross-region inferencing and OpenAI/Anthropic-compatible APIs with native AWS IAM, VPC, and CloudTrail integration
Quick answers
- What is Amazon Bedrock?
- Amazon Bedrock provides API access to 100+ AI models from 18+ providers — including Anthropic Claude, OpenAI GPT-5.6/GPT-6 Astra, Meta Llama 4, Grok 4.7, Amazon Nova, Mistral, and DeepSeek — through AWS-managed infrastructure with no servers to provision. Teams can fine-tune models on private data, build agentic pipelines via Bedrock AgentCore, and ground outputs with managed Knowledge Bases, all within AWS security and compliance boundaries. Pricing is token-based with five billing modes; a free tier is available for experimentation.
- Why does Amazon Bedrock matter?
- Bedrock has become the default enterprise AI infrastructure layer for AWS shops, displacing point solutions by consolidating model access, agent orchestration, RAG, and guardrails into one auditable platform. The April–September 2026 OpenAI partnership — adding GPT-5.6, GPT-6 Astra, and cross-region inferencing — means organizations no longer have to choose between frontier model access and AWS-native security controls. The key risk to watch is cost complexity: five billing modes and per-feature metering (Guardrails, Knowledge Base retrieval, AgentCore transitions) routinely inflate bills 20–30% beyond naive token estimates. Vendor lock-in to AgentCore's orchestration layer is a legitimate architectural concern for teams that may want to run multi-cloud in the future.
- What is Amazon Bedrock best for?
- AWS-native enterprises that need a single control plane for multi-model AI applications with strict data governance, compliance, and agentic workflow requirements
- How much does Amazon Bedrock cost?
- Free tier available for experimentation. On-demand pay-per-token from ~$0.035/M tokens (Amazon Nova Micro) up to $75/M tokens (top-tier frontier models). Batch inference at 50% off on-demand rates. Provisioned throughput and Enterprise (AWS EDP) plans available on request. Separate charges apply for Knowledge Base storage/retrieval, Guardrails, and AgentCore node transitions.
Seen in recent stories
Where Amazon Bedrock showed up in the last 90 days of KeyNews editions.
Accessing OpenAI GPT models on Amazon Bedrock with global cross-Region inference in Thailand, Malaysia, Singapore, Indonesia, and Taiwan
AWS Machine Learning Blog ·
ICYMI: What landed for AI builders in September 2026
AWS Machine Learning Blog ·
How Postman runs Agent Mode for 40 million developers on Amazon Bedrock
AWS Machine Learning Blog ·
Know the terms
Glossary entries that explain what Amazon Bedrock is built on.
Hyperscaler
A cloud computing provider operating at massive scale, primarily Microsoft Azure, Amazon AWS, and Google Cloud. Hyperscalers provide the GPU infrastructure, managed AI services, and global data center networks that power most AI deployments.
RAG (Retrieval-Augmented Generation)
A technique that retrieves relevant documents from an external knowledge base and feeds them to a model alongside the user's query. RAG reduces hallucination and keeps responses grounded in current, factual data.
Agentic orchestration
The architecture pattern of coordinating multiple AI agents to accomplish complex tasks, with a supervisor agent routing work, managing state, and combining results from specialized sub-agents.
Guardrails
Programmatic rules and safety layers that constrain AI model behavior in production. Guardrails can block prompt injection, enforce output formats, prevent policy violations, and ensure brand-safe responses.
Models behind it
The models Amazon Bedrock runs on, offers, or competes with.
More enterprise ai tools
New tools drop every week.
ONE BRIEFING · EVERY FRIDAY · FREE
Free. Unsubscribe anytime.