AWS History and Timeline regarding Amazon Bedrock - Overview, Functions, Features, Summary of Updates, and Introduction

First Published:
Last Updated:

This is another installment in the series that I started with the "AWS History and Timeline - Almost All AWS Services List, Announcements, General Availability(GA)", where I extract features from the history and timeline of AWS services (I've previously written about Amazon S3, AWS Systems Manager, Amazon Route 53, Amazon EventBridge, AWS KMS, Amazon SQS, AWS Lambda, Amazon Cognito, and AWS Machine Learning services in the Japanese edition).

This time, I have created a historical timeline for Amazon Bedrock, the fully managed foundation model service that AWS first announced in April 2023. Since the announcement, Amazon Bedrock has grown from a small set of preview models into a multi-provider platform that now hosts dozens of foundation models, multiple inference modes, knowledge retrieval, guardrails, agent orchestration, and a dedicated runtime for production-grade AI agents called Amazon Bedrock AgentCore.

Just like before, I am summarizing the main features while following the birth of Amazon Bedrock and tracking its feature additions and updates as a Current Overview, Functions, Features of Amazon Bedrock.
I hope these will provide clues as to what has remained the same and what has changed, in addition to the features and concepts of each AWS service.

For readers who want to jump directly to a specific year, the timeline below is grouped chronologically and includes anchors for each year:
  • Timeline 2023 — Announce, GA, and re:Invent 2023 launches (Knowledge Bases, Agents, Guardrails preview, Titan family)
  • Timeline 2024 — Claude 3 family, Llama 3 / 3.1 / 3.2 / 3.3, Mistral, Guardrails GA, Custom Model Import, Converse API, Amazon Nova, Bedrock Marketplace, Prompt Caching
  • Timeline 2025 — Claude 3.7 / Claude 4 / Sonnet 4.5 / Opus 4.5 / Haiku 4.5, DeepSeek, Qwen3, Llama 4, Amazon Nova Sonic / Premier / Nova 2, TwelveLabs, Priority / Flex inference tiers, Amazon Bedrock AgentCore preview and GA
  • Timeline 2026 — Claude Opus 4.6 / Sonnet 4.6 / Opus 4.7, OpenAI models on Bedrock, AgentCore Payments and Agent Registry, latest updates as of 2026-05

Background and Method of Creating Amazon Bedrock Historical Timeline

The reason for creating a historical timeline of Amazon Bedrock this time is that Amazon Bedrock has become the central control plane through which most AWS customers consume foundation models, and the pace of additions has been unusually high since the service went generally available in 2023. New models, new inference modes, new guardrail capabilities, new agent frameworks, and entirely new sub-services (most notably Amazon Bedrock AgentCore) have continued to land at a cadence that makes it difficult to keep a mental map of "when did X arrive on Bedrock?".

Another reason is that since Amazon Bedrock was announced in April 2023, it has been integrated with multiple foundation model providers (Anthropic, Meta, Mistral AI, Cohere, AI21 Labs, Stability AI, Amazon's own Titan and Nova families, DeepSeek, TwelveLabs, Qwen, OpenAI, and increasingly the wider Bedrock Marketplace third-party catalogue) and the supported feature surfaces (Converse API, OpenAI-compatible Chat Completions and Responses APIs, Streaming, Tool Use, Cross-Region Inference, Prompt Caching, Knowledge Bases, Guardrails, Agents, AgentCore, Prompt Flows, Bedrock Studio, Bedrock IDE, Bedrock Marketplace, Bedrock Data Automation, Model Evaluation, Model Distillation, Reinforcement Fine-Tuning) have steadily expanded. Therefore, I wanted to organize the information of Amazon Bedrock with the following approaches.
  • Tracking the history of Amazon Bedrock and organizing the transition of updates
  • Summarizing the feature list and characteristics of Amazon Bedrock
This timeline primarily references the following blogs and document content regarding Amazon Bedrock.
There may be slight variations in the dates on the timeline due to differences in the timing of announcements or article postings in the references used.
The content posted is limited to major features related to the current Amazon Bedrock and necessary for the feature list and overview description.
In other words, please note that the items on this timeline are not all updates to Amazon Bedrock features, but are representative updates that I have picked out.

Amazon Bedrock Historical Timeline (Updates from April 13, 2023)

Now, here is a timeline related to the functions of Amazon Bedrock. As of the time of writing this article, the history of Amazon Bedrock spans about 3 years and 1 month, beginning with the April 2023 announcement of the original preview.

* You can sort the table by clicking on the column name.

Timeline 2023

DateSummary
2023-04-13Amazon Bedrock is announced as a new fully managed generative AI service offering API access to foundation models from AI21 Labs (Jurassic-2), Anthropic (Claude), Stability AI (Stable Diffusion), and Amazon (Titan Text, Titan Embeddings) behind a single managed API, entering limited preview.
AWS Machine Learning Blog
2023-07-25Agents for Amazon Bedrock is announced in preview, a fully managed capability for orchestrating multi-step tasks across foundation models, data sources, and APIs without custom code.
AWS What's New
2023-08-01Anthropic Claude 2 is available on Amazon Bedrock, supporting up to 100,000 token context windows.
AWS What's New
2023-09-13Knowledge Bases for Amazon Bedrock is announced in preview, providing a managed Retrieval-Augmented Generation (RAG) pipeline that connects foundation models to private data in Amazon S3 with vector storage options including Amazon OpenSearch Serverless, Pinecone, and Redis Enterprise Cloud.
AWS What's New
2023-09-28Amazon Bedrock reaches general availability (GA), launching in US East (N. Virginia) and US West (Oregon) with models from AI21 Labs, Anthropic, Cohere, Meta (Llama 2 13B and 70B), Stability AI, and Amazon Titan; includes Provisioned Throughput, HIPAA eligibility, and GDPR compliance.
AWS What's New
2023-09-28Amazon Titan Embeddings reaches general availability with 25+ language support, 8,192 input tokens, and 1,536-dimension output vectors optimized for RAG.
AWS What's New
2023-10-06Amazon Bedrock expands to Asia Pacific (Tokyo), its third region.
AWS What's New
2023-10-19Amazon Bedrock expands to Europe (Frankfurt), its first European region.
AWS What's New
2023-11-28Agents for Amazon Bedrock reaches general availability at AWS re:Invent 2023, enabling agents to plan, call APIs, and retrieve data from company systems in US East (N. Virginia) and US West (Oregon).
AWS What's New
2023-11-28Knowledge Bases for Amazon Bedrock reaches general availability with session context management and source attribution.
AWS What's New
2023-11-28Guardrails for Amazon Bedrock is announced in preview, enabling content filters, denied topics, and other responsible AI policies uniformly across foundation models.
AWS What's New
2023-11-28Amazon Titan Image Generator is announced in preview in Amazon Bedrock.
AWS What's New
2023-11-28Amazon Titan Multimodal Embeddings reaches general availability for embeddings over text and images, powering search, recommendations, and personalization use cases.
AWS What's New
2023-11-28Amazon Titan Text Express and Titan Text Lite reach general availability in Amazon Bedrock.
AWS What's New
2023-11-28Fine-tuning for Meta Llama 2, Cohere Command Light, and Amazon Titan Text becomes available in US East (N. Virginia) and US West (Oregon).
AWS What's New
2023-11-28Continued pre-training for Amazon Titan Text is announced in preview, enabling domain adaptation using unlabeled proprietary data in a secure managed environment.
AWS What's New
2023-11-28Batch inference support launches in Amazon Bedrock, enabling large-scale asynchronous inference jobs over multiple prompts.
AWS What's New
2023-11-29Anthropic Claude 2.1 reaches general availability on Amazon Bedrock, introducing a 200,000-token context window (2x Claude 2.0), reduced hallucination rates, system prompts, and beta tool use for function calling.
AWS What's New
2023-12-21Amazon Bedrock expands to AWS GovCloud (US-West), the first US government cloud region for Bedrock.
AWS What's New

Timeline 2024

DateSummary
2024-03-04Anthropic Claude 3 Sonnet is available on Amazon Bedrock, offering faster inference and higher intelligence than Claude 2.x with a 200K token context window.
AWS What's New
2024-03-14Anthropic Claude 3 Haiku is available on Amazon Bedrock, the most compact and lowest-latency Claude 3 model optimized for near-instant responsiveness and cost efficiency.
AWS What's New
2024-03-26Knowledge Bases for Amazon Bedrock adds Claude 3 Sonnet support via the Retrieve and RetrieveAndGenerate APIs.
AWS What's New
2024-03-28Mistral 7B Instruct and Mixtral 8x7B Instruct reach general availability on Amazon Bedrock, the first Mistral AI models on the platform.
AWS What's New
2024-04-09Amazon Bedrock expands to Asia Pacific (Sydney) and Asia Pacific (Singapore).
AWS What's New
2024-04-09Amazon Bedrock expands to Europe (Paris).
AWS What's New
2024-04-15Meta Llama 3 (8B Instruct and 70B Instruct) is available on Amazon Bedrock in pre-trained and instruction-tuned versions.
AWS What's New
2024-04-16Anthropic Claude 3 Opus is available on Amazon Bedrock, the most advanced and intelligent model in the Claude 3 family with state-of-the-art vision capabilities.
AWS What's New
2024-04-23Guardrails for Amazon Bedrock reaches general availability with four safeguard types (denied topics, content filters, sensitive information filters, word filters) supported for all Bedrock foundation models in English in US East (N. Virginia) and US West (Oregon).
AWS What's New
2024-04-23Model Evaluation for Amazon Bedrock reaches general availability, with automated and human evaluation workflows for comparing foundation models.
AWS What's New
2024-04-23Custom Model Import for Amazon Bedrock is announced in preview, allowing customers to import fine-tuned external models and serve them via Bedrock's serverless inference APIs.
AWS What's New
2024-04-23Amazon Titan Image Generator reaches general availability in Amazon Bedrock.
AWS What's New
2024-04-23Agents for Amazon Bedrock adds simplified agent creation and Return of Control capability, allowing the application to receive intermediate state instead of always invoking Lambda.
AWS What's New
2024-04-23Agents for Amazon Bedrock adds support for Anthropic Claude 3 Haiku and Claude 3 Sonnet.
AWS What's New
2024-04-23Knowledge Bases for Amazon Bedrock adds support for multiple data sources in a single knowledge base.
AWS What's New
2024-04-23Cohere Command R and Command R+ are available on Amazon Bedrock, optimized for long-context RAG, multi-step tool use, and multilingual tasks across 10 languages.
AWS What's New
2024-04-23Amazon Titan Text Embeddings V2 reaches general availability with flexible embedding dimensions (256 / 512 / 1024), 100+ language support, and unit vector normalization.
AWS What's New
2024-05-07Amazon Bedrock Studio is announced in preview, an SSO-enabled web interface for enterprise developers to collaboratively build, evaluate, and share generative AI applications.
AWS What's New
2024-05-07Amazon Titan Text Premier is available on Amazon Bedrock, Amazon's most capable Titan text model.
AWS What's New
2024-05-07Agents for Amazon Bedrock supports the Provisioned Throughput pricing model for predictable performance and cost at scale.
AWS What's New
2024-05-30Amazon Bedrock Converse API is announced, a unified API providing a consistent way to invoke all Bedrock models with multi-turn conversation management and Tool Use (function calling) support for Claude 3, Mistral Large, and Cohere Command R/R+.
AWS What's New
2024-05-30Knowledge Bases for Amazon Bedrock adds Guardrails integration for filtering content and protecting RAG application outputs.
AWS What's New
2024-06-20Anthropic Claude 3.5 Sonnet is available on Amazon Bedrock, surpassing Claude 3 Opus on key benchmarks at one-fifth the cost.
AWS What's New
2024-07-10Amazon Bedrock Prompt Management and Prompt Flows are announced in preview, enabling versioning and evaluation of prompts plus a visual drag-and-drop workflow builder.
AWS What's New
2024-07-10Guardrails for Amazon Bedrock adds contextual grounding checks for hallucination detection and the standalone ApplyGuardrail API that extends safeguards to any self-managed or third-party foundation model.
AWS What's New
2024-07-10Knowledge Bases for Amazon Bedrock adds advanced RAG capabilities including semantic and hierarchical chunking, custom chunking via Lambda, and smart parsing for tabular data in PDFs.
AWS What's New
2024-07-10Knowledge Bases for Amazon Bedrock adds additional data source connectors in preview beyond Amazon S3.
AWS What's New
2024-07-10Agents for Amazon Bedrock adds memory retention capability in preview, maintaining context across multiple sessions.
AWS What's New
2024-07-23Meta Llama 3.1 (8B, 70B, 405B) is available on Amazon Bedrock with a 128K context window and 8-language multilingual support.
AWS What's New
2024-07-26Meta Llama 3.1 405B reaches general availability on Amazon Bedrock, the largest publicly available foundation model from Meta.
AWS What's New
2024-08-21Batch inference pricing reduces to 50% of on-demand price for select foundation models on Amazon Bedrock.
AWS What's New
2024-08-27Amazon Bedrock adds support for cross-region inference, automatically routing requests across AWS regions using inference profiles for improved availability and throughput.
AWS What's New
2024-09-25Meta Llama 3.2 (1B, 3B, 11B Vision, 90B Vision) is available on Amazon Bedrock, adding multimodal vision capabilities to the Llama family.
AWS What's New
2024-09-25Knowledge Bases for Amazon Bedrock adds cross-region inference support for RAG retrieval and generation.
AWS What's New
2024-10-02Amazon Bedrock expands to Asia Pacific (Seoul) and US East (Ohio), bringing commercial region count to 10.
AWS What's New
2024-10-21Amazon Bedrock Custom Model Import reaches general availability, enabling production use of externally trained models via Bedrock's serverless inference APIs.
AWS What's New
2024-10-22Anthropic Claude 3.5 Sonnet v2 and Computer Use beta are available on Amazon Bedrock, with improved coding performance and a public beta of Computer Use for interacting with computer interfaces.
AWS What's New
2024-10-28Meta Llama 3.1 8B and 70B fine-tuning is available on Amazon Bedrock.
AWS What's New
2024-11-04Anthropic Claude 3.5 Haiku is available on Amazon Bedrock, the next-generation Haiku for code suggestions, customer service chatbots, and e-commerce use cases.
AWS What's New
2024-11-07Amazon Bedrock Prompt Management reaches general availability with full lifecycle management, versioning, evaluation, and programmatic retrieval via prompt identifiers in Converse and InvokeModel APIs.
AWS What's New
2024-11-11Amazon Bedrock expands to AWS GovCloud (US-East), the second GovCloud region.
AWS What's New
2024-11-21Prompt Optimization is announced in preview in Amazon Bedrock, automatically rewriting user prompts to better match target foundation-model preferences.
AWS What's New
2024-11-22Amazon Bedrock Flows (Prompt Flows) reaches general availability with real-time workflow execution visibility and Guardrails attachment to Prompt and Knowledge Base nodes.
AWS What's New
2024-11-25InlineAgents launches for Agents for Amazon Bedrock, allowing agent configuration to be defined inline at invocation time rather than requiring pre-created agent resources.
AWS What's New
2024-12-01Guardrails for Amazon Bedrock reduces pricing by up to 85% across all supported regions.
AWS What's New
2024-12-01Amazon Bedrock Rerank API is announced with Amazon Rerank 1.0 and Cohere Rerank 3.5 reranker models for improving relevance of retrieved documents in RAG.
AWS What's New
2024-12-02Latency-optimized inference is announced in public preview on AWS Trainium2 hardware for Anthropic Claude 3.5 Haiku and Meta Llama 3.1 405B/70B.
AWS What's New
2024-12-03Amazon Nova foundation models launch on Amazon Bedrock at AWS re:Invent 2024, including Nova Micro (text only), Nova Lite (multimodal, low cost), Nova Pro (multimodal, balanced), Nova Canvas (image generation), and Nova Reel (video generation); Nova Micro / Lite / Pro support fine-tuning and RAG optimization.
AWS What's New
2024-12-03Amazon Bedrock IDE is announced in preview as part of Amazon SageMaker Unified Studio, a governed collaborative environment for developers to build generative AI applications.
AWS What's New
2024-12-03Multi-Agent Collaboration for Amazon Bedrock is announced in preview, enabling multiple specialized AI agents to coordinate on complex workflows in US East (N. Virginia), US West (Oregon), and Europe (Ireland).
AWS What's New
2024-12-03Guardrails for Amazon Bedrock adds Automated Reasoning checks in preview, providing mathematically verifiable proof that LLM responses are accurate with 99% claimed accuracy.
AWS What's New
2024-12-04Amazon Bedrock Data Automation (BDA) is announced in preview, automating insight extraction from unstructured multimodal content (documents, images, videos, audio) into structured JSON.
AWS What's New
2024-12-04Prompt Caching is announced in preview on Amazon Bedrock, claiming up to 90% cost reduction and 85% latency reduction by caching frequently used prompt prefixes for Claude 3.5 Haiku, Claude 3.5 Sonnet v2, and Amazon Nova models.
AWS What's New
2024-12-04Intelligent Prompt Routing is announced in preview on Amazon Bedrock, automatically routing requests within a model family based on predicted complexity.
AWS What's New
2024-12-04Knowledge Bases for Amazon Bedrock adds GraphRAG support in preview using Amazon Neptune Analytics for graph-based retrieval.
AWS What's New
2024-12-04Knowledge Bases for Amazon Bedrock adds structured data retrieval support, enabling natural-language queries over relational databases.
AWS What's New
2024-12-04Knowledge Bases for Amazon Bedrock adds RAG Evaluation in preview, measuring retrieval accuracy and generation faithfulness.
AWS What's New
2024-12-04Amazon Bedrock Marketplace launches with 100+ models, a unified catalog of publicly available and proprietary foundation models accessible via Bedrock APIs and compatible with Agents, Knowledge Bases, and Guardrails.
AWS What's New
2024-12-19Meta Llama 3.3 70B is available on Amazon Bedrock, delivering similar performance to Llama 3.1 405B at a fraction of computational cost.
AWS What's New
2024-12-23Amazon Bedrock Agents, Flows, and Knowledge Bases add latency-optimized model support via Trainium2-powered inference.
AWS What's New

Timeline 2025

DateSummary
2025-02-24Anthropic Claude 3.7 Sonnet is available on Amazon Bedrock, a hybrid reasoning model supporting standard mode and extended thinking mode with adjustable reasoning budgets; first FedRAMP High / DoD CC SRG IL4/5 approved Claude in GovCloud.
AWS What's New
2025-03-03Amazon Bedrock Data Automation reaches general availability in US West (Oregon) and US East (N. Virginia) with improved document accuracy, video summarization, logo detection for 35,000+ brands, cross-region inference, AWS KMS CMK encryption, and AWS PrivateLink support.
AWS What's New
2025-03-07Knowledge Bases for Amazon Bedrock GraphRAG reaches general availability in all regions where both Bedrock Knowledge Bases and Amazon Neptune Analytics are available.
AWS What's New
2025-03-10Multi-Agent Collaboration for Amazon Bedrock reaches general availability with expanded region support.
AWS What's New
2025-03-10DeepSeek-R1 is available as a fully managed model on Amazon Bedrock, with AWS announcing it as the first major cloud provider to offer DeepSeek-R1 as a fully managed and serverless foundation model in US East (N. Virginia), US East (Ohio), and US West (Oregon) via cross-region inference.
AWS What's New
2025-03-20Amazon Bedrock Model Evaluation LLM-as-a-Judge reaches general availability, with LLM-based automated assessment of helpfulness, accuracy, and coherence.
AWS What's New
2025-03-20Amazon Bedrock RAG Evaluation reaches general availability, measuring retrieval precision, faithfulness, and answer relevance.
AWS What's New
2025-04-07Prompt Caching reaches general availability on Amazon Bedrock for supported models including Claude 3.5 / 3.7 and Amazon Nova.
AWS What's New
2025-04-08Amazon Nova Sonic speech-to-speech model launches on Amazon Bedrock, unifying speech understanding and generation in a single foundation model with function calling and RAG support, plus a new bidirectional streaming API.
AWS What's New
2025-04-22Intelligent Prompt Routing reaches general availability on Amazon Bedrock, routing within Anthropic Claude and Meta Llama families based on complexity prediction.
AWS What's New
2025-04-23Prompt Optimization reaches general availability on Amazon Bedrock.
AWS What's New
2025-04-29Meta Llama 4 (Scout 17B and Maverick 17B) is available on Amazon Bedrock, the first Llama 4 models with mixture-of-experts architecture, multimodal (text+image) in 12 languages, with Scout offering 10M token context and Maverick 1M context with 128 experts.
AWS What's New
2025-04-30Amazon Nova Premier launches on Amazon Bedrock, Amazon's most capable multimodal model with 1M token context, 87.4% MMLU, and 82.0% Math500; also the most capable teacher model for Bedrock Model Distillation.
AWS What's New
2025-05-01Amazon Bedrock Model Distillation reaches general availability, with teacher models (Nova Premier, Claude 3.5 Sonnet v2, Llama 3.3 70B) and student models (Nova Pro, Llama 3.2 1B / 3B).
AWS What's New
2025-05-22Anthropic Claude 4 (Opus 4 and Sonnet 4) is available on Amazon Bedrock, both hybrid reasoning models with standard and extended thinking modes; Opus 4 is Anthropic's most powerful model and top coding model per benchmarks.
AWS What's New
2025-07-15TwelveLabs Marengo 2.7 and Pegasus 1.2 are available on Amazon Bedrock, the first TwelveLabs models on the platform for video embedding (Marengo) and video-language modeling (Pegasus).
AWS What's New
2025-07-16Amazon Bedrock AgentCore is announced in preview, a comprehensive enterprise-grade platform with seven modular services: Runtime (session isolation, 8-hour workloads), Memory (short / long-term across sessions), Gateway (API-to-MCP-tool conversion), Browser (cloud-based browser runtime), Code Interpreter (sandboxed code execution), Identity (IdP integration with Cognito / Entra / Okta), and Observability (end-to-end agent visibility).
AWS What's New
2025-08-05Anthropic Claude Opus 4.1 is available on Amazon Bedrock, a refined Opus 4 model with enhanced agentic and complex reasoning capabilities.
AWS What's New
2025-09-18DeepSeek-V3.1 is available as fully managed on Amazon Bedrock with switchable thinking / non-thinking modes, enhanced tool calling, and reduced hallucinations vs. previous DeepSeek versions.
AWS What's New
2025-09-18Qwen3 models are available as fully managed on Amazon Bedrock, including Qwen3-235B-A22B and Qwen3-32B across US, Asia Pacific, Europe, and South America regions.
AWS What's New
2025-09-29Anthropic Claude Sonnet 4.5 is available on Amazon Bedrock, Anthropic's mid-tier model with enhanced capabilities for complex agents, coding tasks, and enterprise workflows, plus improved context management.
AWS What's New
2025-10-13Amazon Bedrock AgentCore reaches general availability, with all seven services (Runtime, Memory, Gateway, Browser, Code Interpreter, Identity, Observability) GA in nine AWS regions, full VPC / AWS PrivateLink / CloudFormation / tagging support, A2A protocol on Runtime, and Gateway connectivity to existing MCP servers.
AWS What's New
2025-10-15Anthropic Claude Haiku 4.5 is available on Amazon Bedrock, delivering near-frontier performance matching Claude Sonnet 4's capabilities in coding, computer use, and agent tasks at lower cost and faster speeds, with a 200K token context window and reasoning support.
AWS What's New
2025-11-18Amazon Bedrock Priority and Flex inference service tiers launch, with Priority for mission-critical real-time and Flex for non-interactive batch workloads at discounted pricing.
AWS What's New
2025-11-19Amazon Bedrock Custom Model Import adds support for OpenAI GPT OSS models, enabling import and serverless serving of OpenAI's open-weight models.
AWS What's New
2025-11-24Anthropic Claude Opus 4.5 is available on Amazon Bedrock, setting new standards for coding, agentic workflows, computer use, and office tasks at one-third the cost of prior Opus models; introduces tool search and tool use examples via Bedrock API.
AWS What's New
2025-11-30Multimodal retrieval for Bedrock Knowledge Bases reaches general availability, allowing retrieval of images, charts, and other non-text content.
AWS What's New
2025-12-02Amazon Bedrock adds 18 fully managed open weight models, the largest single expansion to date spanning multiple model families and use cases.
AWS What's New
2025-12-02AgentCore Policy and Evaluations are announced in preview: Policy uses natural-language rules auto-converted to Cedar for real-time tool-call interception via Gateway; Evaluations provides 13 built-in evaluators; Memory adds episodic memory; Runtime adds bidirectional streaming for voice agents.
AWS What's New
2025-12-02Amazon Nova 2 foundation models launch on Amazon Bedrock, the next generation of Nova text and multimodal models.
AWS What's New
2025-12-02Amazon Nova 2 Sonic launches for real-time conversational AI on Amazon Bedrock, the next generation of the Nova Sonic speech-to-speech model.
AWS What's New
2025-12-04Amazon Bedrock adds support for the OpenAI Responses API via the Mantle inference engine, enabling asynchronous inference, stateful conversation management, and OpenAI SDK compatibility through a base URL change.
AWS What's New

Timeline 2026

DateSummary
2026-01-26Amazon Bedrock adds 1-hour duration support for prompt caching, extending the maximum cached prefix duration.
AWS What's New
2026-01-29Amazon Bedrock adds server-side custom tools via the Responses API, allowing agents to call AWS-provided tools (notes, tasks) or custom Lambda functions without client round-trips.
AWS What's New
2026-02-05Anthropic Claude Opus 4.6 is available on Amazon Bedrock, Anthropic's flagship model for coding, enterprise agents, and professional work; supports both 200K and 1M context tokens (1M in preview).
AWS What's New
2026-02-10Amazon Bedrock adds 6 fully managed open weight models: DeepSeek V3.2, MiniMax M2.1, GLM 4.7, GLM 4.7 Flash, Kimi K2.5, and Qwen3 Coder Next.
AWS What's New
2026-02-12Amazon Bedrock expands AWS PrivateLink support for OpenAI-compatible API endpoints (Chat Completions and Responses APIs) across 14+ commercial regions.
AWS What's New
2026-02-17Anthropic Claude Sonnet 4.6 is available on Amazon Bedrock, a direct upgrade of Sonnet 4.5 with improved coding, computer use, long-context reasoning, and agent planning, with 1M token context.
AWS What's New
2026-02-17Amazon Bedrock Reinforcement Fine-Tuning (RFT) adds support for open-weight models with OpenAI-compatible APIs for OpenAI GPT-OSS and Qwen, with fine-tuned models deployable immediately via Chat Completions and Responses APIs.
AWS What's New
2026-02-26Amazon Bedrock announces the OpenAI-compatible Projects API (Mantle), providing IAM-based access control and cost tagging for OpenAI-compatible Chat Completions and Responses API usage.
AWS What's New
2026-02-27Amazon Bedrock batch inference supports the Converse API format, simplifying large-scale asynchronous inference.
AWS What's New
2026-03-13Amazon Bedrock AgentCore Runtime adds support for the AG-UI protocol, enabling responsive real-time agent-to-user experiences; AgentCore now supports MCP, A2A, and AG-UI protocols.
AWS What's New
2026-03-17Amazon Bedrock expands to Asia Pacific (New Zealand), a new commercial region.
AWS What's New
2026-04-03Guardrails for Amazon Bedrock cross-account safeguards reach general availability, applying Guardrail configurations across multiple AWS accounts for centralized governance.
AWS What's New
2026-04-09AWS Agent Registry launches in preview via Amazon Bedrock AgentCore, a private governed catalog and discovery layer for agents, tools, skills, MCP servers, and custom resources within an organization.
AWS What's New
2026-04-16Anthropic Claude Opus 4.7 is available on Amazon Bedrock.
AWS What's New
2026-04-22Amazon Bedrock AgentCore adds a CLI, managed harness, and skills for coding assistants: the AgentCore CLI supports infrastructure-as-code deployment with CDK (Terraform coming); the managed harness provides filesystem persistence for session suspend / resume; AgentCore skills are available for Kiro Power with Claude Code / Codex / Cursor support coming.
AWS What's New
2026-04-28Amazon Bedrock launches OpenAI models, Codex, and Managed Agents in limited preview, with OpenAI frontier models available via Bedrock APIs with full IAM / PrivateLink / Guardrails / CloudTrail controls; Codex coding agent via CLI / VS Code; Managed Agents powered by OpenAI harness running in the AWS environment.
AWS What's New
2026-04-30Amazon Bedrock AgentCore launches performance optimization capabilities in preview: Recommendations analyzes production traces to generate optimized system prompts and tool descriptions; batch evaluations validate against test cases; A/B tests validate against live traffic with statistical-significance reporting; all changes require explicit approval.
AWS What's New
2026-05-05Amazon Bedrock AgentCore is available in AWS GovCloud (US-West) for workloads with elevated compliance requirements.
AWS What's New
2026-05-07Amazon Bedrock AgentCore adds the Payments capability in preview, enabling AI agents to autonomously pay for APIs, MCP servers, web content, and other agents using the x402 protocol; built with Coinbase and Stripe, handling wallet auth, stablecoin payment, and deterministic spending limits.
AWS What's New

Current Overview, Functions, Features of Amazon Bedrock

From here, I will explain in detail the main features of Amazon Bedrock as of 2026-05.

Amazon Bedrock is a fully managed service that exposes a unified API for inference against a curated set of foundation models from multiple model providers (Anthropic, Meta, Mistral AI, Cohere, AI21 Labs, Stability AI, Amazon's own Titan and Nova families, DeepSeek, TwelveLabs, Qwen, OpenAI, and the broader Amazon Bedrock Marketplace catalog). Amazon Bedrock differs from Amazon SageMaker JumpStart in that it never asks the customer to provision compute capacity; inference is a serverless API call, and capacity-based pricing is only opted into when the customer explicitly buys Provisioned Throughput.

Amazon Bedrock streamlines generative AI application development by providing a single API surface across providers, while also offering higher-level building blocks such as Knowledge Bases (managed RAG), Guardrails (policy-based content moderation), Agents (multi-step orchestration with tool calls), Amazon Bedrock AgentCore (production-grade agent infrastructure), Prompt Flows and Prompt Management, Bedrock Studio, Bedrock IDE in Amazon SageMaker Unified Studio, Bedrock Marketplace, Bedrock Data Automation, Model Evaluation, and Model Distillation. Along with these features, it integrates with other AWS services through IAM, AWS PrivateLink, CloudTrail, CloudWatch, KMS, and VPC, allowing for rapid development of secure generative AI applications.

Amazon Bedrock Service Overview

Amazon Bedrock occupies the highest abstraction layer for generative AI on AWS:
  • Amazon Bedrock: API-based access to managed foundation models and managed AI building blocks (RAG, guardrails, agents). No infrastructure to manage.
  • Amazon SageMaker JumpStart: Self-managed deployment of pretrained models into SageMaker-managed endpoints. The customer owns the endpoint and pays per hour for the instance.
  • Amazon SageMaker AI: Lower-level ML platform for full custom training and deployment.
Customers typically start at the Bedrock layer for stateless inference and move down to JumpStart or SageMaker AI only when they need custom-trained models, specialized hardware sizing, or workload patterns that benefit from owning the endpoint.

Amazon Bedrock Conceptual Diagram

From here, I will explain the main features and characteristics of Amazon Bedrock, but before that, I will show the following conceptual diagram of Amazon Bedrock to make it easier to imagine the overall picture of Amazon Bedrock.

Amazon Bedrock Platform Overview
Amazon Bedrock Platform Overview

This diagram illustrates how a user application interacts with Amazon Bedrock through two complementary APIs (the high-level Converse API and the low-level InvokeModel API), and how the request fans out through orchestration and tooling components (Agents, AgentCore, Knowledge Bases, Guardrails, Prompt Flows / Prompt Management, Bedrock Data Automation) into one of several inference modes (On-Demand, Provisioned Throughput, Cross-Region Inference Profiles, Prompt Caching / Intelligent Prompt Routing, Batch Inference) and finally lands on a specific foundation model hosted on Bedrock or the Bedrock Marketplace.

For more concrete API examples and code patterns, see the companion article Amazon Bedrock Basic Information and API Examples. For the model catalog as of a specific point in time, see Amazon Bedrock Models As of 2024.

Foundation Models Available on Amazon Bedrock

Amazon Bedrock hosts foundation models from the following providers as of 2026-05. Exact model availability varies by AWS region; consult the Bedrock console "Model access" page for the authoritative current list.
  • Anthropic Claude family: Claude Instant (legacy), Claude 2 / 2.1, Claude 3 Opus / Sonnet / Haiku, Claude 3.5 Sonnet (v1 and v2), Claude 3.5 Haiku, Claude 3.7 Sonnet (extended thinking), Claude Opus 4 / Sonnet 4, Claude Opus 4.1, Claude Sonnet 4.5, Claude Opus 4.5, Claude Haiku 4.5, Claude Opus 4.6, Claude Sonnet 4.6, Claude Opus 4.7.
  • Meta Llama family: Llama 2 (13B / 70B Chat), Llama 3 (8B / 70B Instruct), Llama 3.1 (8B / 70B / 405B), Llama 3.2 (1B / 3B text-only and 11B / 90B vision), Llama 3.3 (70B), Llama 4 (Scout 17B, Maverick 17B).
  • Mistral AI: Mistral 7B Instruct, Mixtral 8x7B Instruct, Mistral Large.
  • Cohere: Command, Command Light, Command R, Command R+, Embed English, Embed Multilingual, Cohere Rerank 3.5.
  • AI21 Labs: Jurassic-2 Ultra / Mid (legacy), Jamba family.
  • Stability AI: Stable Diffusion XL 1.0 and successive Stable Image models.
  • Amazon Titan: Titan Text Lite / Express / Premier, Titan Text Embeddings V1 / V2, Titan Image Generator, Titan Multimodal Embeddings, Amazon Rerank 1.0.
  • Amazon Nova: Nova Micro (text only), Nova Lite (low-cost multimodal), Nova Pro (multimodal), Nova Canvas (image generation), Nova Reel (video generation), Nova Sonic (speech-to-speech), Nova Premier, and the Nova 2 generation announced at re:Invent 2025 including Nova 2 Sonic.
  • DeepSeek: DeepSeek-R1 (fully managed since March 2025), DeepSeek-V3.1, DeepSeek V3.2.
  • TwelveLabs: Marengo 2.7 (video embedding), Pegasus 1.2 (video language model).
  • Qwen: Qwen3-235B-A22B, Qwen3-32B, Qwen3 Coder Next, and additional Qwen3 variants.
  • OpenAI (limited preview since 2026-04-28): GPT OSS open-weight models via Custom Model Import; OpenAI frontier models, Codex, and Managed Agents via Bedrock APIs.
  • Bedrock Marketplace (third-party): 100+ additional foundation models from independent model providers, deployed onto Amazon SageMaker endpoints in the customer's AWS account and accessible under a single Bedrock control plane.

Amazon Bedrock Inference Modes

Amazon Bedrock supports the following inference modes:
  • On-Demand: Pay-per-token API calls with no capacity reservation. Quotas are enforced as tokens-per-minute (TPM) and requests-per-minute (RPM) limits per account, per region, per model.
  • Provisioned Throughput: Reserved capacity expressed as Model Units (MUs). Used for predictable workloads or for models that require fine-tuning. Quotas are guaranteed; pricing is hourly per MU.
  • Cross-Region Inference Profiles: A virtual model ID that automatically routes on-demand calls across regions within a continental group, raising the effective on-demand quota and improving availability without code changes.
  • Batch Inference: Asynchronous job-based inference for large input sets, with discounted per-token cost (50% of on-demand for many models). Supports the Converse API message format as of 2026-02-27.
  • Prompt Caching: Cache common prompt prefixes (system prompts, retrieved context, few-shot examples) at the inference layer for cheaper and faster repeated calls; up to 1-hour cache duration (2026-01-26). Supported across the Claude 3.5+ family and Amazon Nova family.
  • Intelligent Prompt Routing: Automatically dispatches each request to the cheapest model in a configured family that is predicted to meet a quality bar.
  • Priority and Flex service tiers (2025-11-18): Priority for mission-critical real-time interactions; Flex for non-interactive batch workloads at discounted pricing.

Knowledge Bases for Amazon Bedrock

Knowledge Bases for Amazon Bedrock is a managed Retrieval-Augmented Generation (RAG) pipeline. The customer points Bedrock at an Amazon S3 bucket (or other supported data source), and Bedrock chunks the content, generates embeddings with a configured embedding model, and stores them in a configured vector store (Amazon OpenSearch Serverless, Amazon Aurora PostgreSQL with pgvector, Pinecone, Redis Enterprise Cloud, or MongoDB Atlas). At query time, Bedrock retrieves the top-k relevant chunks, optionally re-ranks them via the Rerank API (Amazon Rerank 1.0 or Cohere Rerank 3.5), and supplies them as context to the chosen foundation model.

Knowledge Bases also supports structured data retrieval (natural-language SQL against Amazon Redshift) and GraphRAG (graph-based retrieval over Amazon Neptune Analytics), with GraphRAG reaching general availability on 2025-03-07 and structured data retrieval added on 2024-12-04. Multimodal retrieval for images, charts, and other non-text content reached GA on 2025-11-30.

Guardrails for Amazon Bedrock

Guardrails for Amazon Bedrock applies safety, privacy, and topic policies uniformly across foundation models. A single configured Guardrail can be reused across multiple models. Guardrails cover:
  • Topic denial policies
  • Content filters (hate, insults, sexual, violence, misconduct, prompt attacks)
  • Word filters (custom blocked words and regex patterns)
  • Sensitive information filters (PII detection, redaction, anonymization)
  • Contextual grounding checks (hallucination detection against supplied context, added 2024-07-10)
  • Automated Reasoning checks (formal verification against a customer policy, added 2024-12-03 in preview)
  • Cross-account safeguards (GA on 2026-04-03)
Guardrails can be invoked inline with an inference call or independently via the ApplyGuardrail API, which makes them usable with non-Bedrock models as a separate moderation layer. For complementary network-layer defenses such as WAF rate-based rules and prompt injection patterns, see the companion article AWS WAF Generative AI Prompt Injection Patterns.

Agents for Amazon Bedrock

Agents for Amazon Bedrock orchestrate multi-step tasks by:
  1. Decomposing a user request into a plan
  2. Calling Lambda-backed action groups (tools), API Gateway endpoints, or built-in Knowledge Bases
  3. Reading and re-planning based on intermediate observations
  4. Returning a final natural-language response together with tool-call traces
Multi-Agent Collaboration (announced in preview at re:Invent 2024, GA on 2025-03-10) extends Agents with a supervisor pattern: a supervisor agent coordinates multiple specialist agents within a single Bedrock-managed workflow. InlineAgents (2024-11-25) allow agent configuration to be defined inline at invocation time without pre-created agent resources.

Amazon Bedrock AgentCore

Amazon Bedrock AgentCore (announced in preview on 2025-07-16, GA on 2025-10-13) packages production-grade agent infrastructure into seven services:
  • AgentCore Runtime: Long-running secure microVM sandboxes for executing agent code, with session isolation, up to 8-hour workloads, bidirectional streaming for voice agents (2025-12-02), and protocol support for MCP, A2A, and AG-UI (2026-03-13).
  • AgentCore Memory: Two-layer memory store (short-term conversational state + long-term semantic memory) with managed eviction and episodic memory (added 2025-12-02).
  • AgentCore Identity: Per-user token vault that handles OAuth flows on behalf of the agent, integrating with Amazon Cognito, Microsoft Entra ID, and Okta as identity providers.
  • AgentCore Gateway: HTTP-to-MCP tool federation layer that turns any HTTP API or AWS Lambda function into a Model Context Protocol (MCP) tool consumable by AgentCore Runtime or any MCP-compatible agent. Gateway also connects to existing MCP servers (added 2025-10-13).
  • AgentCore Code Interpreter: Sandboxed code execution environment with a managed Python runtime for tool-augmented agent reasoning.
  • AgentCore Browser: Cloud-based headless browser runtime for agent-driven web automation.
  • AgentCore Observability: OpenTelemetry-native tracing and span hierarchy for agent invocations, integrated with Amazon CloudWatch.
Additional AgentCore capabilities added after GA include AgentCore Policy and Evaluations in preview (2025-12-02), AgentCore Payments capability in preview using the x402 protocol (2026-05-07), an AWS Agent Registry as a governed catalog of agents and tools (2026-04-09), and an AgentCore CLI with managed harness for session suspend / resume and IaC-based deployment with CDK (2026-04-22). Performance optimization capabilities (Recommendations, batch evaluations, A/B tests) entered preview on 2026-04-30.

For a full implementation walkthrough, see the companion AgentCore series on this site:

Prompt Flows, Prompt Management, Bedrock Studio, and Bedrock IDE

  • Prompt Management (GA 2024-11-07): Versioned prompt library inside Bedrock, decoupling prompt text from application code; programmatically retrievable via Converse and InvokeModel APIs by prompt identifier.
  • Prompt Flows / Amazon Bedrock Flows (GA 2024-11-22): Visual builder for chaining prompts, Lambda functions, S3 readers, Knowledge Bases, Agents, and Guardrails. Each flow becomes its own invokable resource.
  • Prompt Optimization (preview 2024-11-21, GA 2025-04-23): Automatically rewrites user prompts to maximize model-specific response quality.
  • Amazon Bedrock Studio (preview 2024-05-07): SSO-enabled web interface for enterprise developers to collaboratively build, evaluate, and share generative AI applications.
  • Amazon Bedrock IDE (preview 2024-12-03, part of Amazon SageMaker Unified Studio): A governed collaborative environment for developers to build and tailor generative AI applications.

Amazon Bedrock Marketplace

Amazon Bedrock Marketplace (announced at re:Invent 2024) surfaces 100+ additional foundation models from independent model providers inside the Bedrock console. Models are deployed onto Amazon SageMaker endpoints in the customer's AWS account under a single Bedrock control plane, so the same Converse API and InvokeModel API can target Marketplace models alongside Bedrock-native models.

Amazon Bedrock Data Automation

Amazon Bedrock Data Automation (preview 2024-12-04, GA 2025-03-03) provides a managed pipeline that extracts insights from unstructured multimodal content (documents, images, videos, audio) into structured JSON output. Customers define a "project" with a target output schema; Bedrock Data Automation orchestrates the appropriate combination of OCR, transcription, video keyframe extraction, logo detection (35,000+ brands), and foundation-model summarization to populate that schema. GA shipped with cross-region inference, AWS KMS CMK encryption, and AWS PrivateLink support.

Amazon Bedrock Model Evaluation and Model Distillation

  • Model Evaluation (GA 2024-04-23): Automated and human evaluation workflows. The LLM-as-a-Judge variant (GA 2025-03-20) uses an LLM to assess helpfulness, accuracy, and coherence. RAG Evaluation (GA 2025-03-20) measures retrieval precision, faithfulness, and answer relevance.
  • Model Distillation (GA 2025-05-01): Trains a smaller "student" model on outputs from a larger "teacher" model. Teacher models include Amazon Nova Premier, Claude 3.5 Sonnet v2, and Llama 3.3 70B; student models include Nova Pro and Llama 3.2 1B / 3B; distilled models are claimed up to 500% faster and 75% cheaper with <2% accuracy loss.
  • Reinforcement Fine-Tuning (RFT) (2026-02-17): Extended to OpenAI GPT-OSS and Qwen with OpenAI-compatible fine-tuning APIs; fine-tuned models deployable via Chat Completions and Responses APIs.

OpenAI-compatible APIs (Mantle) on Amazon Bedrock

Amazon Bedrock's Mantle inference engine adds OpenAI-compatible endpoints. The OpenAI Responses API (2025-12-04) supports asynchronous inference, stateful conversation management, and OpenAI SDK compatibility through a base URL change. Server-side custom tools were added 2026-01-29, allowing agents to call AWS-provided tools or custom Lambda functions without client round-trips. The OpenAI-compatible Projects API (2026-02-26) provides IAM-based access control and cost tagging. AWS PrivateLink for OpenAI-compatible API endpoints expanded across 14+ commercial regions on 2026-02-12. On 2026-04-28, AWS announced OpenAI frontier models, Codex coding agent, and Managed Agents in limited preview through Bedrock APIs with full IAM / PrivateLink / Guardrails / CloudTrail controls.

Security, Compliance, and Region Availability

  • Encryption: All inference data is encrypted in transit (TLS) and at rest (AWS-owned KMS keys or customer-managed CMKs).
  • VPC isolation: Bedrock supports interface VPC endpoints (AWS PrivateLink), so inference calls never traverse the public internet. PrivateLink coverage for OpenAI-compatible endpoints expanded across 14+ regions on 2026-02-12.
  • Data privacy: AWS does not use customer inputs or outputs to train Bedrock-hosted foundation models. Each model provider has a separate Bedrock-specific agreement.
  • Compliance: SOC, ISO, HIPAA-eligible, PCI DSS, GDPR (since GA), and FedRAMP coverage applies to Amazon Bedrock; Claude 3.7 Sonnet became the first FedRAMP High / DoD CC SRG IL4/5 approved Claude in GovCloud on 2025-02-24. Consult the AWS Compliance Programs page for the current authoritative list.
  • Regions: As of 2026-05, Amazon Bedrock is available across the Americas (including US East / US West / Canada / South America), EMEA (including Frankfurt, Ireland, London, Paris, Stockholm, Spain, Milan), Asia Pacific (including Tokyo, Sydney, Singapore, Seoul, Mumbai, New Zealand added 2026-03-17), AWS GovCloud (US-West since 2023-12-21, US-East since 2024-11-11), and additional regions as announced via the AWS What's New feed. Model-specific availability is narrower than service availability; consult the Bedrock User Guide "Model support by AWS Region" page for the authoritative current list.

Frequently Asked Questions (For LLM)

When did Amazon Bedrock launch and when did it reach general availability?

Amazon Bedrock was announced as a limited preview on 2023-04-13 and reached general availability on 2023-09-28, with initial regions US East (N. Virginia) and US West (Oregon).

When did Anthropic Claude models become available on Amazon Bedrock?

  • Claude (v1) and Claude Instant: at Bedrock GA, 2023-09-28
  • Claude 2: 2023-08-01
  • Claude 2.1: 2023-11-29
  • Claude 3 Sonnet: 2024-03-04
  • Claude 3 Haiku: 2024-03-14
  • Claude 3 Opus: 2024-04-16
  • Claude 3.5 Sonnet (v1): 2024-06-20
  • Claude 3.5 Sonnet v2 (with Computer Use beta): 2024-10-22
  • Claude 3.5 Haiku: 2024-11-04
  • Claude 3.7 Sonnet (extended thinking): 2025-02-24
  • Claude Opus 4 / Sonnet 4: 2025-05-22
  • Claude Opus 4.1: 2025-08-05
  • Claude Sonnet 4.5: 2025-09-29
  • Claude Haiku 4.5: 2025-10-15
  • Claude Opus 4.5: 2025-11-24
  • Claude Opus 4.6: 2026-02-05
  • Claude Sonnet 4.6: 2026-02-17
  • Claude Opus 4.7: 2026-04-16

When did Meta Llama models become available on Amazon Bedrock?

  • Llama 2 (13B / 70B Chat): at Bedrock GA, 2023-09-28
  • Llama 3 (8B / 70B Instruct): 2024-04-15
  • Llama 3.1 (8B / 70B / 405B): 2024-07-23 (405B GA 2024-07-26)
  • Llama 3.2 (1B / 3B / 11B Vision / 90B Vision): 2024-09-25
  • Llama 3.3 (70B): 2024-12-19
  • Llama 4 (Scout 17B, Maverick 17B): 2025-04-29

When did Mistral / Cohere / AI21 / Stability AI / Amazon Nova / DeepSeek / TwelveLabs / Qwen / OpenAI models become available on Amazon Bedrock?

  • Mistral 7B Instruct and Mixtral 8x7B Instruct GA: 2024-03-28
  • Cohere Command and Embed: at Bedrock GA, 2023-09-28
  • Cohere Command R / R+: 2024-04-23
  • Amazon Rerank 1.0 and Cohere Rerank 3.5: 2024-12-01
  • AI21 Jurassic-2: at Bedrock GA, 2023-09-28
  • Stability AI Stable Diffusion XL: at Bedrock GA, 2023-09-28
  • Amazon Titan Text / Embeddings: at Bedrock GA, 2023-09-28
  • Amazon Titan Image Generator GA: 2024-04-23
  • Amazon Nova (Micro / Lite / Pro / Canvas / Reel): announced at AWS re:Invent 2024 on 2024-12-03
  • Amazon Nova Sonic (speech-to-speech): 2025-04-08
  • Amazon Nova Premier: 2025-04-30
  • Amazon Nova 2 and Nova 2 Sonic: 2025-12-02
  • DeepSeek-R1 (fully managed): 2025-03-10
  • DeepSeek-V3.1: 2025-09-18
  • DeepSeek V3.2 and 5 other open-weight models: 2026-02-10
  • TwelveLabs Marengo 2.7 and Pegasus 1.2: 2025-07-15
  • Qwen3 models: 2025-09-18
  • OpenAI GPT OSS (via Custom Model Import): 2025-11-19
  • OpenAI frontier models, Codex, Managed Agents (limited preview): 2026-04-28

When did Amazon Bedrock Guardrails reach general availability?

Guardrails for Amazon Bedrock was announced in preview at AWS re:Invent 2023 on 2023-11-28 and reached general availability on 2024-04-23. Contextual grounding checks and the independent ApplyGuardrail API were added on 2024-07-10. Automated Reasoning checks entered preview on 2024-12-03. Cross-account safeguards reached GA on 2026-04-03. Pricing was reduced by up to 85% on 2024-12-01.

When did Knowledge Bases for Amazon Bedrock launch and reach GA?

Knowledge Bases for Amazon Bedrock was announced in preview on 2023-09-13 and reached general availability at AWS re:Invent 2023 on 2023-11-28. GraphRAG reached GA on 2025-03-07. Multimodal retrieval reached GA on 2025-11-30. Structured data retrieval was added 2024-12-04. The Rerank API was added 2024-12-01.

When did Agents for Amazon Bedrock reach GA and when was Multi-Agent Collaboration added?

Agents for Amazon Bedrock was announced in preview on 2023-07-25 and reached general availability at AWS re:Invent 2023 on 2023-11-28. Multi-Agent Collaboration was announced in preview at re:Invent 2024 on 2024-12-03 and reached general availability on 2025-03-10. InlineAgents launched on 2024-11-25.

When did Amazon Bedrock AgentCore launch and reach GA?

Amazon Bedrock AgentCore was announced in preview on 2025-07-16 and reached general availability on 2025-10-13 across nine AWS regions. AgentCore consists of seven services: Runtime, Memory, Identity, Gateway, Code Interpreter, Browser, and Observability. AgentCore Policy and Evaluations entered preview on 2025-12-02. AWS Agent Registry entered preview on 2026-04-09, the AgentCore CLI / managed harness / coding-assistant skills launched on 2026-04-22, performance optimization capabilities entered preview on 2026-04-30, and AgentCore Payments entered preview on 2026-05-07. AgentCore in AWS GovCloud (US-West) launched on 2026-05-05.

When did Bedrock add Prompt Caching, Cross-Region Inference, Tool Use, the Converse API, and OpenAI-compatible APIs?

  • Streaming responses: supported from Bedrock GA (2023-09-28) onward for compatible models
  • Batch inference: launched 2023-11-28
  • Converse API (unified API with standardized Tool Use / function calling): 2024-05-30
  • Cross-Region Inference Profiles: 2024-08-27
  • Batch inference 50% pricing: 2024-08-21
  • Prompt Management and Prompt Flows preview: 2024-07-10; Prompt Management GA 2024-11-07; Flows GA 2024-11-22
  • Prompt Caching: announced preview at re:Invent 2024 (2024-12-04); reached GA 2025-04-07; 1-hour cache duration added 2026-01-26
  • Intelligent Prompt Routing: preview 2024-12-04, GA 2025-04-22
  • OpenAI Responses API support via Mantle: 2025-12-04
  • Server-side custom tools via Responses API: 2026-01-29
  • PrivateLink for OpenAI-compatible API endpoints: 2026-02-12
  • Projects API (Mantle): 2026-02-26
  • Batch inference Converse API format: 2026-02-27
  • AG-UI protocol on AgentCore Runtime: 2026-03-13

Which AWS regions support Amazon Bedrock as of the latest snapshot?

As of 2026-05, Amazon Bedrock is available in major AWS commercial regions across the Americas, EMEA, and Asia Pacific (including New Zealand, added 2026-03-17), plus AWS GovCloud (US-West since 2023-12-21) and AWS GovCloud (US-East since 2024-11-11). Model-specific availability is narrower than service availability; consult the Bedrock User Guide "Model support by AWS Region" page for the authoritative current list.

What is the difference between Amazon Bedrock and Amazon SageMaker JumpStart?

Amazon Bedrock provides API-based, serverless inference against managed foundation models with no infrastructure to manage. Amazon SageMaker JumpStart deploys pretrained models onto SageMaker-managed endpoints that the customer owns and pays for hourly. Bedrock is the higher-abstraction surface; JumpStart is used when the customer needs endpoint-level control, specialized instance sizing, or hardware not supported by Bedrock.

How is Amazon Bedrock AgentCore different from Agents for Amazon Bedrock?

Agents for Amazon Bedrock is the original orchestration layer for multi-step task execution inside the Bedrock control plane, designed for the "supervisor + action groups" pattern with declarative configuration. Amazon Bedrock AgentCore is a separate set of production-grade infrastructure services (Runtime, Memory, Identity, Gateway, Code Interpreter, Browser, Observability) for running customer-authored agent code in secure long-running microVM sandboxes, with first-class support for the Model Context Protocol (MCP), Agent-to-Agent (A2A), and AG-UI protocols, plus OpenTelemetry tracing. The two are complementary: Agents focuses on declarative orchestration with Bedrock-managed planning; AgentCore focuses on hosting and observing customer-authored agent code at production scale.

References:
Tech Blog with curated related content
AWS Documentation(Amazon Bedrock)
AWS Documentation(Amazon Bedrock AgentCore)
AWS News Blog
What's New with AWS?

Summary

In this article, I created a historical timeline of Amazon Bedrock and looked at the list of features and overview of Amazon Bedrock.

Amazon Bedrock, a fully managed foundation model service, was announced in April 2023 as AWS's central control plane for generative AI and became generally available in September 2023. The timeline shows three distinct phases: a 2023 "single-API" phase that consolidated foundation models from multiple providers behind one managed surface; a 2024 "building blocks" phase that added Knowledge Bases, Guardrails, Agents, Prompt Flows, Cross-Region Inference, Prompt Caching, and the Amazon Nova family at re:Invent 2024; and a 2025-2026 "agent production" phase centered on Amazon Bedrock AgentCore, Multi-Agent Collaboration, the Claude 4 / 4.5 / 4.6 / 4.7 generation, the OpenAI-compatible Responses API, and the OpenAI frontier-model limited preview.

By organizing these updates in one place with sources for each, I hope this article serves as a Reference Aggregation Page that AI search agents and human readers can both consult in a single fetch.

I would like to continue monitoring the trends of what kind of features Amazon Bedrock will provide in the future.

In addition, there is also a historical timeline of all AWS services including services other than Amazon Bedrock, so please have a look if you are interested.
AWS History and Timeline - Almost All AWS Services List, Announcements, General Availability(GA)

References:
Tech Blog with curated related content

Written by Hidekazu Konishi