AWS AI Updates: June 17, 2026

1. Grok 4.3 from xAI Now Available in Amazon Bedrock

Amazon Bedrock. xAI’s reasoning-first Grok 4.3 model is now available in Amazon Bedrock with configurable reasoning effort across four levels (none, low, medium, and high), giving developers control over latency and token cost per request. AWS positions the model for strong tool use and instruction following in agentic workflows, with target use cases spanning customer support, web development, case law research, and financial document Q&A. Grok 4.3 runs on Mantle, AWS’s new inference engine in Bedrock, which supports tool calling, structured output, and response streaming. Source

2. Amazon Bedrock Guardrails Adds Resourceless InvokeGuardrailChecks API for Agents

Amazon Bedrock. Bedrock Guardrails now offers InvokeGuardrailChecks, a resourceless API that applies individual safeguards at any step of an agentic application without creating or version-managing a guardrail resource. It runs in detect-only mode and returns numeric severity and confidence scores for customizable thresholds, supporting content filters, prompt attack detection (jailbreak, prompt injection, and prompt leakage as independent checks), and sensitive information filters. The per-request, granular control targets multi-step agent loops where uniform protection is impractical, and it is available in US East (N. Virginia, Ohio), US West (Oregon), Europe (London, Stockholm), and Asia Pacific (Tokyo, Sydney). Source

3. AWS Transform Adds Model-to-Model Migration Assessment for Generative AI

AWS Transform. AWS Transform can now scan codebases and generate a migration plan for moving generative AI workloads from third-party providers to Amazon Bedrock, including cost comparisons and production-ready code changes. It supports migrations from OpenAI, Google Gemini, direct Anthropic SDK usage, and open-source models via LiteLLM or Ollama, and it handles framework-wrapped patterns such as LangChain and LlamaIndex plus agentic architectures including CrewAI and LangGraph. The tool adds tiered model routing recommendations, prompt caching analysis, and model lifecycle awareness, and is available in all regions where AWS Transform is offered at no additional charge. Source

4. Amazon S3 Vectors Now Returns Up to 10,000 Results Per Query

Amazon S3 Vectors. S3 Vectors now returns up to 10,000 similarity search results per query, a 100x increase over the prior 100-result limit, enabling multi-stage retrieval pipelines that apply reranking, aggregation, or deduplication before producing a final result set. Results are paginated so applications can begin processing the first page while later pages are still being retrieved, and developers specify the desired topK value in QueryVectors API requests using the latest AWS SDK. A data-returned fee applies to larger result sets, with the first 512 KB per query offered at no charge, and the capability is available in all regions where S3 Vectors runs. Source

5. Amazon S3 Adds Annotations to Give AI Agents Business Context

Amazon S3. S3 now supports annotations, custom metadata in JSON, XML, or YAML attached directly to objects at up to 1GB per object, letting AI agents and analytics tools understand data without a separate metadata system. Annotations move with the object during copy and replication, are removed when the object is deleted, and inherit the object’s durability and consistency properties. Through S3 Metadata, annotations surface in read-only Apache Iceberg tables queryable with Amazon Athena, and they can be searched in natural language using agents in Amazon SageMaker Unified Studio or any IDE with the S3 Tables MCP server. Annotations are available in all AWS Regions, including the AWS China Regions. Source