AWS Bedrock User Guide · 2026-09-01
Added model cards for <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-fable-5-1.html">Claude Fable 5.1</a> and <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-mythos-5-1.html">Claude Mythos 5.1</a>. Updated the Adaptive thinking, Extended thinking, Tool use, and Prompt caching pages. Added a new <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/claude-messages-thinking-block-binding.html">Thinking block binding (beta)</a> topic.
Open original
AWS Bedrock User Guide · 2026-08-24
Corrected the Reasoning entry on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-sonnet-5.html">Claude Sonnet 5</a> model card, which incorrectly stated that adaptive thinking is always on and cannot be disabled. On Claude Sonnet 5, adaptive thinking is on by default — including for requests that omit the <code class="code">thinking</code> field — and can be turned off by passing <code class="code">"thinking": <span>{</span>"type": "disabled"}</code>. Added Claude Sonnet 5 to the supported models table on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/inference-anthropic-claude-messages-adaptive-thinking.html">Adaptive thinking</a> topic, and added guidance covering the default-on behavior change from Claude Sonnet 4.6, the fact that only the explicit <code class="code">disabled</code> thinking type turns thinking off (omitting the <code class="code">thinking</code> field selects adaptive thinking, and <code class="code">budget_tokens</code> is not supported on this model), that thinking tokens are billed as output tokens, and the need to revisit <code class="code">max_tokens</code> for workloads migrated from Claude Sonnet 4.6.
Open original
AWS Bedrock User Guide · 2026-08-19
Fixed the Anthropic Messages API Python sample on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html">Overview</a> page, which previously used <code class="code">anthropic.Anthropic()</code> with no base URL and failed with a validation exception. Every Anthropic Messages API Python sample now uses the plain <code class="code">Anthropic</code> client from the Anthropic SDK, targets the <code class="code">bedrock-runtime</code> endpoint's <code class="code">/anthropic</code> route directly with <code class="code">base_url="https://bedrock-runtime.<span>{</span>region}.amazonaws.com/anthropic"</code>, and authenticates with a short-term Amazon Bedrock bearer token generated by <a href="https://pypi.org/project/aws-bedrock-token-generator/" rel="noopener noreferrer" target="_blank">aws-bedrock-token-generator</a>'s <code class="code">provide_token()</code>. Updated the Sample Code section of every Anthropic model card that supports <code class="code">bedrock-runtime</code> (Claude Opus 5, Opus 4.8, Opus 4.7, Sonnet 5, Haiku 4.5, and Fable 5) to use this pattern with the model's Global inference ID, and updated the corresponding <code class="code">pip install</code> line to <code class="code">pip install -U anthropic aws-bedrock-token-generator</code>. Updated the same sample on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/getting-started.html">Quickstart</a> page, the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/inference-messages-api.html">Messages API</a> topic, and the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/inference-anthropic-claude-messages-tool-use.html">Tool use</a> reference topic. Reordered the code sample tabs on the Messages API topic to lead with <code class="code">bedrock-runtime</code> variants (Anthropic SDK, boto3, AWS CLI, curl) and moved the <code class="code">bedrock-mantle</code> curl tab to the end. Repointed the <code class="code">OPENAI_BASE_URL</code> environment variable on 49 non-Anthropic model cards (OpenAI, Mistral AI, DeepSeek, Google, Meta, Qwen, MiniMax, NVIDIA, Moonshot AI, Z.AI, Writer, xAI) from the <code class="code">bedrock-mantle</code> <code class="code">/v1</code> route to the <code class="code">bedrock-runtime</code> <code class="code">/openai/v1</code> route (<code class="code">https://bedrock-runtime.<span>{</span>region}.amazonaws.com/openai/v1</code>) so the Responses API and Chat Completions API samples in the Sample Code section run against <code class="code">bedrock-runtime</code> by default. The <code class="code">bedrock-mantle</code> endpoint continues to be fully supported and pages that document mantle-only features (Web Search, Projects, Workspaces, Chat Completions on mantle, the mantle Responses API page) still show <code class="code">bedrock-mantle</code> URLs.
Open original
AWS Bedrock User Guide · 2026-08-15
Follow-up updates continuing the <code class="code">bedrock-runtime</code> positioning work. Reordered the code sample tabs on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/getting-started.html">Quickstart</a> page to lead with the Messages API, followed by the Responses, Chat Completions, Converse, and Invoke APIs. The Messages API tab now targets the <code class="code">bedrock-runtime</code> endpoint using the <code class="code">AnthropicBedrock</code> client (AWS SigV4 auth), so all Quickstart tabs consistently use <code class="code">bedrock-runtime</code>. Updated the Quickstart code samples to use Claude Opus 5 for Anthropic and GPT-5.6 Sol for OpenAI. Added a suggestion callout under the Model Details section of every model card that supports <code class="code">bedrock-runtime</code>, recommending the <code class="code">bedrock-runtime</code> endpoint for new applications. Added Daybreak Red: GPT-5.6 Cyber and Daybreak Blue: GPT-5.6 Sol to the OpenAI row of the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html">Models at a glance</a> summary table.
Open original
AWS Bedrock User Guide · 2026-08-15
Updated the endpoint and API guidance across the Amazon Bedrock User Guide to recommend the <code class="code">bedrock-runtime</code> endpoint for new applications, including for OpenAI- and Anthropic-compatible workloads. Changes include reordered quickstart tabs and code samples on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html">Overview</a> and <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/getting-started.html">Quickstart</a> pages, reordered endpoint and API guidance on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/endpoints.html">Endpoints</a>, <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/apis.html">APIs</a>, and <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/build.html">Build</a> pages, reordered Programmatic Access sections in model cards, retitled the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/bedrock-mantle.html">Responses API</a> and <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/inference-chat-completions-mantle.html">Chat Completions API</a> pages that document the <code class="code">bedrock-mantle</code> variants, and added cross-reference tips on the <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/projects.html">Projects</a> and <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/workspaces.html">Workspaces</a> pages. The <code class="code">bedrock-mantle</code> endpoint continues to be fully supported and existing applications that use it do not need to change.
Open original
AWS Bedrock User Guide · 2026-08-14
Agentic retrieval on fully managed Amazon Bedrock knowledge bases now integrates with <a href="https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/memory.html">Amazon Bedrock AgentCore Memory</a>. Use the <code class="code">memoryConfiguration</code> field on <code class="code">AgenticRetrieveStream</code> to configure memory. You can restore an earlier session from short-term memory, let the agent retrieve memory records from long-term memory, or both. For more information, see <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/kb-test-agentic-retrieve.html">Use agentic retrieval to query a knowledge base</a>.
Open original
AWS Bedrock User Guide · 2026-08-11
Amazon Bedrock now supports cost allocation by IAM principal (users and roles) for inference requests made through the <code class="code">bedrock-mantle</code> endpoint, extending the capability previously available for <code class="code">bedrock-runtime</code>. Tag IAM users and roles with attributes such as team, project, or cost center, activate them as cost allocation tags, and analyze <code class="code">bedrock-mantle</code> inference costs by those tags in AWS Cost Explorer or at the line-item level in AWS Cost and Usage Report 2.0 (CUR 2.0). Available in all AWS Regions where the <code class="code">bedrock-mantle</code> endpoint is available. For more information, see <a href="https://docs.aws.amazon.com/bedrock/latest/userguide/cost-mgmt-iam-principal-tracking.html">IAM principal attribution in Amazon Bedrock</a>.
Open original
AWS Bedrock User Guide · 2026-07-30
Amazon Bedrock Agents (now Amazon Bedrock Agents Classic) is no longer open to new customers. For capabilities similar to Bedrock Agents Classic, explore <a href="https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/agentcore-get-started-cli.html">Amazon Bedrock AgentCore</a>. Existing customers can continue to use the service as normal. For more information, see
<a href="https://docs.aws.amazon.com/bedrock/latest/userguide/agents-classic-maintenance-mode.html">Amazon Bedrock Agents Classic availability change</a>.
Open original
AWS Bedrock User Guide · 2026-06-30
Amazon Bedrock Agents (launched November 2023) is now Amazon Bedrock Agents Classic and will no longer be open to new customers starting on July 30, 2026. If you would like to use Bedrock Agents Classic, sign up prior to that date. Existing customers can continue to use the service as normal. For more information, see
<a href="https://docs.aws.amazon.com/bedrock/latest/userguide/agents-classic-maintenance-mode.html">Amazon Bedrock Agents Classic maintenance mode</a>.
Open original
AWS Bedrock User Guide · 2026-05-27
Documented the per-model input and output token quotas that govern inference traffic to the <code class="code">bedrock-mantle</code> endpoint, including the default values for Anthropic Claude Opus 4.7, the request evaluation order, the differences from <code class="code">bedrock-runtime</code> quotas, and how to request quota increases through AWS Support.
Open original
AWS Bedrock User Guide · 2026-05-27
Added a dedicated page that describes the per-model token quotas that govern inference traffic to the <code class="code">bedrock-runtime</code> endpoint and points to related quotas for custom inference profiles, batch inference, and Provisioned Throughput.
Open original
AWS Bedrock User Guide · 2026-05-27
Clarified RPM quota behavior across Amazon Bedrock endpoints. The <code class="code">bedrock-mantle</code> endpoint does not expose per-model RPM quotas; throttling is governed by token-based quotas. On the <code class="code">bedrock-runtime</code> endpoint, RPM quotas are model-specific — some models, including Anthropic Claude Opus 4.7 and Claude Opus 4.8, are not subject to an RPM quota. For models that do have an RPM quota, see the Service Quotas console.
Open original
AWS Bedrock User Guide · 2026-05-26
Documented how to count input tokens through the Anthropic <code class="code">count_tokens</code> path on the <code class="code">bedrock-mantle</code> endpoint, including for Claude models that are not supported by <a href="https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_CountTokens.html">CountTokens</a> on <code class="code">bedrock-runtime</code>.
Open original
AWS Bedrock User Guide · 2026-03-02
Amazon Bedrock Knowledge Bases now offers fully managed knowledge bases, where Amazon Bedrock manages the storage, indexing, and retrieval infrastructure for you. Fully managed knowledge bases also support agentic retrieval with the AgenticRetrieveStream API, which uses a foundation model to decompose complex queries into sub-queries and iteratively retrieve relevant information.
Open original
AWS Bedrock User Guide · 2026-02-27
You can now use the Converse API format for batch inference input data. When creating a batch inference job, set the model invocation type to Converse to use a consistent request format across models.
Open original
AWS Bedrock User Guide · 2026-02-17
Added support for fine-tuning open-weight models using OpenAI-compatible APIs. You can now create, monitor, and manage reinforcement fine-tuning jobs for open-weight models through familiar OpenAI SDK endpoints, including Files API, fine-tuning jobs APIs, and inference APIs.
Open original
AWS Bedrock User Guide · 2025-12-31
Updated list of models supported for priority and flex service tiers
Open original
AWS Bedrock User Guide · 2025-12-18
Amazon Bedrock Data Automation now supports blueprint instruction optimization for documents. Improve Data Automation custom output accuracy by providing example content assets with ground truth labels, enabling you to achieve production-ready accuracy for documents in minutes without model training.
Open original
AWS Bedrock User Guide · 2025-12-03
Added new reinforcement fine-tuning to improve foundation model performance through feedback signals.
Open original
AWS Bedrock User Guide · 2025-12-03
Amazon Bedrock now supports OpenAI compatible API endpoints, including Responses API and Chat Completions API. These endpoints enable asynchronous inference for long-running workloads, stateful conversation management without manual history passing, and simplified tool use integration for agentic workflows. Migrate existing applications with minimal code changes by updating your base URL and API key.
Open original