Amazon Web Services

Amazon Web Services researchers working with Booking.com combined knowledge distillation and the Medusa-2 framework to train a 1.1-billion-parameter TinyLlama that matched the quality of a 70-billion-parameter Llama-3.1 teacher. According to the authors, it ran 10 to 20 times as fast depending on the use case.

Audible's technology team describes a controllable text-to-speech system that identifies which character speaks each line with 97% accuracy in English and 94% in Spanish. Rendering distinct character voices improved dialogue differentiation by 3.7 points on a MUSHRA test, with statistically significant listening-experience gains for eight of ten voices.

AWS has published a reference architecture for a voice travel concierge that pairs Amazon Nova 2.5 Sonic with Bedrock AgentCore runtime, AgentCore Gateway over MCP, and Bedrock Knowledge Bases. It deploys with a single AWS CDK script and runs against a sample airline backend with synthetic data. The post does not disclose pricing, context window, or benchmark figures for Nova 2.5 Sonic.

Amazon Bedrock now offers GLM 5.3, Z.ai's 753B-parameter mixture-of-experts model, through managed APIs with cross-Region inference, prompt caching and service tiers. Access is limited to eligible enterprise customers. Pricing and context window were not disclosed in AWS's announcement.

Amazon SageMaker AI has released the aws-ai-ml skill, distributed through the Agent Toolkit for AWS. It lets MCP-compatible coding agents such as Kiro, Claude Code and Codex benchmark endpoints, recommend deployment configurations and generate executable SageMaker Python SDK v3 code.

AWS has published the Quick Resource Migrator, a sample MCP server on Amazon Bedrock AgentCore that promotes Amazon Quick resources between AWS accounts in a single tool call. It is idempotent, never deletes from the target, and writes versioned S3 backups before every update.

AWS published a walkthrough for connecting Claude Desktop on Amazon Bedrock to a managed, MCP-compatible Web Search capability through Amazon Bedrock AgentCore Gateway. According to AWS, the search is backed by an Amazon web index spanning tens of billions of documents, and query traffic stays within AWS infrastructure. Web Search is available in three AWS Regions; pricing is not disclosed in the post.

Amazon Web Services has released Strands Decider 2B, an open-source model that chooses among pre-decided options and returns a confidence score instead of generating text. It is inspired by TypeSafe's Jev and is small enough to run locally. Amazon says it briefly topped the Jevbench ranking for models of its size.

AWS published a reference implementation for ambient agents on Amazon Bedrock AgentCore. S3 uploads or scheduled events create jobs that an agent runs, pausing for human input through a single ask_human tool. Each agent turn is capped at the 15-minute Lambda timeout.

Taiwan retail platform uniopen adapted Amazon Nova 2 Lite to its two-axis moderation policy using LoRA supervised fine-tuning in Amazon SageMaker AI, plus a prompt-format change. According to AWS, Per Behavior Macro F1 rose from 0.5852 to 0.8550 and Subject Type Macro F1 from 0.4162 to 0.8491, both above production targets.

AWS published a reference architecture showing how to combine structured data extraction with Bedrock AgentCore, dual Claude models, and Amazon Quick to answer portfolio-wide questions that standard RAG systems get wrong. The design uses Claude Sonnet 4.6 for extraction and Claude Haiku 4.5 for independent verification, with Amazon Textract as a deterministic tiebreaker.

AWS published a deployment guide for running Alibaba's Qwen3-TTS-12Hz-1.7B-Base voice cloning model as a real-time SageMaker inference endpoint. The model clones a speaker's voice from a short audio clip and generates speech in 10 languages, including cross-lingual cloning, without retraining.

Aderant built a serverless ticket triage system on Amazon Nova Lite that reviewed 109 tickets in its first 2.5 weeks with roughly 96% routing accuracy. The company estimates the system recovers 8-14 engineering hours per week at under $30 in total monthly operating cost.

Amazon Bedrock now supports the open source coding agent OpenCode paired with open weight models including Moonshot AI's Kimi K3, OpenAI's GPT-OSS 120B, and NVIDIA's Nemotron 3 Super 120B. The setup keeps inference inside a customer's AWS account with per-token pricing instead of per-seat subscriptions.

Amazon rolled out 13 new SageMaker AI inference capabilities in 2026 across managed endpoints and HyperPod Inference, spanning automated benchmarking, instance-pool fallback, OpenAI-compatible APIs, and container caching. AWS claims container caching cut endpoint startup latency by 51% and an inference-recommendation feature doubled GPT-OSS-20B throughput at equal latency.

Amazon has released an updated Bedrock AgentCore runtime that holds cold start latency at roughly 2 seconds regardless of container image size, versus up to 30 seconds on the previous version. The update also changes memory billing to track real-time usage instead of peak allocation.

Wood Mackenzie built APEX (Agentic Platform for Energy eXperience) on Amazon Bedrock AgentCore to give three separate applications a shared runtime for identity, guardrails, memory, and scaling instead of each rebuilding the same infrastructure. The company says 88% of its internal AI proofs-of-concept never reach wide deployment, a gap it attributes to architecture rather than model quality.

Amazon Bedrock's prompt caching feature can cut input token costs by up to 90% on cache hits by storing repeated context like documents, system prompts, and tool definitions. AWS outlines six implementation patterns and pricing details, including a 25% premium for cache writes and 90% discount on cache reads.

AWS published a technical walkthrough for building MCP Apps—interactive HTML widgets rendered inside AI hosts like ChatGPT and Claude—using Amazon Bedrock AgentCore's runtime and Gateway components. The reference implementation, a unicorn rental app, demonstrates host-agnostic rich UI delivered through a single MCP server.

AWS has launched model caching for Amazon SageMaker Inference on HyperPod, pre-loading model weights and container images onto cluster nodes to cut inference pod startup times from tens of minutes to seconds. The feature targets large models like DeepSeek-R1, which previously took 30+ minutes to become servable during scale-out events.