From AWS Machine Learning Blog · 40 stories
NVIDIA Launches Revenue-Sharing Model for AI Infrastructure and Agent Toolkit
NVIDIA has introduced a revenue-sharing model for AI cloud partners to access its infrastructure more affordably, enabling startups to pay a percentage of revenue in addition to hardware costs. Additionally, NVIDIA released an Agent Toolkit to facilitate the creation of specialized AI systems within business workflows. These initiatives aim to expand NVIDIA's AI technology reach and revenue sources.
Strategic Frameworks and Systems Vital for Successful AI Integration in Enterprises
AI's integration in enterprises is moving beyond model development to focus on creating robust systems for execution and governance. This shift highlights the importance of developing adaptable frameworks to support AI's role across various functions such as finance, HR, and operations. It reflects a broader industry trend where the focus is on building the necessary infrastructure to ensure AI's ongoing, safe, and productive incorporation into real-world workflows, addressing the current challenges and limitations.
OpenAI's GPT-6 Astra Achieves High Scores on ARC-AGI-3 Benchmark
OpenAI has released GPT-6 Astra, its latest AI model, which achieved a 99.9% score on the ARC-AGI-3 benchmark using a provider adapter harness. This marks a significant improvement over its predecessor, GPT-5.6 Sol, which scored 7.8%, and demonstrates the model's ability to navigate unfamiliar interactive environments and create symbolic world models.
New AI Models for Long-Horizon Coding Tasks Introduced
Several AI models aimed at long-horizon tasks in coding and robotics have been released. GLM-5.2 by Hugging Face extends support for coding-agent scenarios with a 1 million token context. Cognition's SWE-1.7 enhances long-horizon asynchronous tasks with reinforcement learning. Xiaomi-Robotics-1 combines vast pre-training data for improved robotics capabilities. These releases highlight advances in scaling and reasoning capabilities.
Meta Introduces Hybrid Asset Classification for Privacy-Aware Infrastructure
Meta has unveiled a hybrid asset classification strategy using large language models (LLMs) to handle ambiguous data in privacy-aware infrastructure while maintaining deterministic rules for enforcement. This method addresses the complexities of AI-native products with varied data inputs, ensuring compliance and effective data governance. It is a response to the challenges posed by the increasing speed and scale of AI innovations, and the approach aims to better manage privacy controls for evolving AI products.
U.S. Greenlights Public Rollout of OpenAI's GPT-5.6 Amid Regulatory Controls
The U.S. government has approved OpenAI's GPT-5.6 models for public release on July 9, ending a period of limited access due to regulatory scrutiny. The launch of these models, including Sol, Terra, and Luna, follows compliance with federal cybersecurity reviews intended to manage AI model rollouts. This episode highlights the tension between advancing AI capabilities and the increasing regulatory oversight.
Amazon Bedrock Enhances AI Capabilities with Security and Operational Features
Amazon Bedrock has introduced several updates to improve the security and operational management of AI applications, emphasizing capabilities for multi-tenant AI, data retention policies, and compliance with US government standards. Key features include resource-based policies, managed entitlements for model subscriptions, zero data retention enforcement, and AI model support in AWS GovCloud. These advancements aim to streamline AI adoption across diverse sectors while maintaining security and governance standards.
Claude Mythos 5 AI Cybersecurity Capabilities Expanded, $35M Fund for Open-Source Security
Claude Mythos 5, an AI model for cybersecurity, is now available in Claude Security and will integrate into partner tools. The company also launched a $35 million fund to support open-source software security and plans to expand its Cyber Verification Program. These actions aim to broaden access to advanced AI for defensive cybersecurity while maintaining safeguards against misuse.
Challenges in AI Token Costs and Efficiency Revealed
AI model pricing based on tokens has been criticized for being misleading due to varying tokenization methods. DeepSeek's price cut on its V4-Pro model exemplifies the complexity as lower token rates don't guarantee cost savings. Researchers highlight solutions like AI harnesses that optimize token usage, offering cost-effective alternatives.
Amazon Quick Enhances Efficiency for Finance, Sales, and Supply Chain Management
Amazon Quick, a generative AI assistant, enhances efficiency for AWS Finance, sales teams, Tradeshift, and supply chain operations. It streamlines data preparation, administrative tasks, and analytics processes, allowing organizations to improve productivity and response times across various sectors.
Anthropic Releases Claude Fable 5.2 and Mythos 5.2 with Speed, Cost, and Capability Upgrades
Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, updated AI models designed for complex technical challenges. These models offer up to 30% faster output generation, reduced pricing to $8 per million input tokens, and improvements in coding, reasoning, and vision. Claude Fable 5.2 is generally available, while Claude Mythos 5.2 is accessible via verification programs for cybersecurity and life sciences.
SpaceXAI Releases Grok Bot AI Agent and Grok 4.6 Model, Completes Cursor Acquisition
SpaceXAI, which recently completed its acquisition of AI coding company Cursor, has launched Grok Bot, an AI agent for Mac, iOS, Windows, and Linux, designed to automate tasks across applications. Concurrently, the company released Grok 4.6, an updated AI model that scores 61 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Sol and offering competitive pricing for long-running agents, coding, and knowledge work.
Anthropic Releases Claude Opus 5, Offering Near Fable 5 Performance at Half the Cost
Anthropic has released Claude Opus 5, a new AI model that approaches the intelligence of its flagship Fable 5 model but at half the price. Opus 5 is now the default model for Claude Max subscribers and the strongest available for Claude Pro users, offering improved performance in coding and knowledge work while maintaining the token cost of its predecessor, Opus 4.8.
Alibaba Releases Qwen3.8-Max and Qwen3.8-27B AI Models, Including Open Weights
Alibaba has released its Qwen3.8-Max and Qwen3.8-27B AI models. Qwen3.8-Max is a multimodal model with 2.4 trillion parameters and a 1 million token context window, while Qwen3.8-27B is a 27-billion-parameter version with native vision-language understanding. The company plans to release open weights for both models, making Qwen-Max-class capabilities available to the open-source community.
Amazon SageMaker AI Integrates with MLflow for Enhanced Model Deployment and Monitoring
Amazon SageMaker AI has integrated with MLflow to provide a unified tracking interface for AI inference, benchmark, and model monitoring. This supports both generative AI models and discriminative ML models, simplifying model deployment and improving ongoing accuracy by detecting data drift. The integration is designed to reduce setup time and enhance workflow reproducibility in machine learning applications.
Model Context Protocol (MCP) 2026-07-28 Specification Released, Adopting Stateless Core
The Model Context Protocol (MCP) has released its 2026-07-28 specification, transitioning from a bidirectional stateful protocol to a request/response stateless core. This update, the largest revision since its launch, aims to improve reliability and scalability for MCP servers, addressing a highly requested developer feature. Major SDKs, including TypeScript, Python, and C#, have been updated to support the new specification.
SpaceXAI Launches Cost-Effective AI Model Grok 4.5, Challenging Rivals
SpaceXAI released Grok 4.5, a new AI model built in collaboration with Cursor, focusing on coding and engineering tasks. The model offers improved efficiency and lower costs, presenting a competitive alternative to existing AI models. Grok 4.5's release introduces a pricing strategy that undercuts rivals, influencing the AI market dynamics among enterprise users and developers.
AWS Introduces TorchNeuron for Direct PyTorch Execution on Trainium AI Chips
Amazon Web Services (AWS) has launched TorchNeuron, a new PyTorch plugin that enables direct execution of AI applications on AWS Trainium AI chips. This integration allows developers to run PyTorch models on Trainium with minimal configuration changes, supporting eager mode, `torch.compile`, and native distributed APIs, and expanding out-of-the-box support for Hugging Face text generation models.
Anthropic Releases Economical AI Model Claude Sonnet 5 for Enhanced Agentic Tasks
Anthropic has unveiled Claude Sonnet 5, an advanced AI model designed for agentic capabilities, available on AWS. This model provides substantial improvements in planning, tool use, and coding over previous versions at a lower cost, aiming to compete with top-tier AI models.
Inscribe Uses Amazon Bedrock for Rapid Document Fraud Detection
Inscribe utilizes Amazon Bedrock to enhance its document fraud detection, reducing verification time to under 90 seconds. This advancement addresses the surge in AI-generated fraud and the pressing need for financial institutions to maintain accuracy amidst increasing application volumes.
Amazon SageMaker HyperPod Adds Silent Data Corruption Detection and Recovery
Amazon SageMaker HyperPod now includes new detection and recovery mechanisms for silent data corruption (SDC) in distributed training. These features address SDC, which occurs when hardware returns incorrect but plausible numbers without error, by integrating checks directly into the PyTorch DistributedDataParallel training loop.
Kimi K3: New 2.8 Trillion-Parameter Open AI Model with Vision Capabilities Released
Kimi has introduced Kimi K3, a 2.8 trillion-parameter AI model with built-in vision capabilities and a 1-million-token context window. This open-source model, while not as powerful as proprietary models, shows competitive performance across various benchmarks, including a strong showing against Fable 5. Kimi K3 offers new potential for AI applications but faces cost and speed challenges.
Anthropic Launches Claude Apps Gateway for AWS and Google Cloud
Anthropic has launched the Claude apps gateway for both AWS and Google Cloud. This tool provides enterprises with centralized management of Claude Code, enhancing control over access, costs, and policy adherence. By streamlining developer credential management and spend tracking, the gateway eases administrative burdens for organizations operating at scale.
Moonshot AI's Kimi K3.1 Pro Highspeed LLM Now Available on Amazon Bedrock
Moonshot AI has made its Kimi K3.1 Pro Highspeed large language model available on Amazon Bedrock. This new model, part of the K3.1 series, offers improved performance for long-horizon agentic work, professional tasks, and coding, along with 2-3 times faster output generation compared to Kimi K3.1 Pro.
Amazon Polly Launches Voice Cloning for Seven Languages
Amazon Polly has made its voice cloning feature generally available as a gated release, allowing users to create a synthetic voice from a few minutes of audio recordings. This feature supports seven languages and runs on Polly's generative engine, providing natural-sounding cloned voices for various applications.
Amazon Bedrock Managed Knowledge Base Adds Tabular Search for Data Analysis
Amazon Bedrock Managed Knowledge Base now supports tabular search, enabling users to extract answers from CSV, XLSX, and XLS files through agentic retrieval. This feature allows for calculations across table rows and combines results with document facts to answer complex analytical questions. It addresses challenges of traditional RAG with tables by processing entire files in an isolated sandbox.
OpenFold3-preview for Biomolecular Structure Prediction Now Deployable on Amazon SageMaker JumpStart
Amazon SageMaker JumpStart now supports the deployment of OpenFold3-preview, an open-source biomolecular structure prediction model developed by the OpenFold Consortium and AlQuraishi Lab. This integration allows users to predict 3D structures of proteins, DNA, RNA, and their complexes within their AWS accounts, addressing challenges associated with managing GPU infrastructure and complex software stacks for such models.
SGLang-Neuron Publicly Available for Frontier AI Model Deployment on AWS Trainium
Amazon announced the public availability of SGLang-Neuron, enabling users to serve AI models and generate reinforcement learning rollouts with SGLang on AWS Trainium. This integration allows developers to use their existing SGLang workflows and APIs, extending support to AWS's AI chips for training and inference.
Semantic Diffing for Formal Policies Using Anti-Unification Engine
An open-source engine now computes semantic differences between formal policies using anti-unification on an e-graph. This method identifies behavioral changes by accounting for algebraic laws and domain rules, unlike textual diffs that highlight syntactic variations. The engine helps in reviewing autoformalized policies and comparing different formalizations generated by large language models.
Amazon Quick Introduces Smart Approvals for AI Agent Action Oversight
Amazon Quick launched Smart Approvals, a feature that automates routine AI agent actions while preserving human oversight for critical decisions. This system evaluates proposed actions against existing security controls and user requests, automatically approving safe actions or flagging others for human review. Smart Approvals aims to reduce approval fatigue by focusing human attention on necessary judgments without altering existing permissions or mandatory approvals.
Circle K reduces incident triage time to under 15 minutes using Amazon Bedrock
Circle K implemented an automated incident triage system using Amazon Bedrock, reducing their mean triage time from 45-60 minutes to under 15 minutes. This system uses an agentic AI architecture to autonomously investigate incidents and provide structured summaries to engineers via Slack, allowing for faster remediation.
Worldwide Express Automates Freight Exception Handling with Amazon Bedrock, Reducing Unresolved Cases by 75%
Worldwide Express implemented a multi-agent AI solution using Amazon Bedrock to automate freight billing exception handling within its transportation management system. This automation led to a 75% reduction in active exceptions, a four-week improvement in time-to-cash, and $1 million in annual savings.
AWS Ray Train DLCs and EFA on EKS simplify distributed PyTorch training
AWS provides Ray Train Deep Learning Containers (DLCs) that integrate necessary components for distributed PyTorch training on Amazon EKS, including CUDA, PyTorch, NCCL, EFA, and the AWS OFI NCCL plugin. This offering simplifies the setup and maintenance of distributed training environments by providing pre-built and tested container images, removing the need for users to manage complex dependencies.
Property Finder fixes evaluation bias to improve AI models on Amazon Bedrock
Property Finder discovered significant evaluation bias in their AI models, where 96.8% of evaluation labels were inherited from a single model's predictions instead of human verification. By building unbiased ground truth and using evolutionary prompt optimization on Amazon Bedrock, they achieved higher accuracy and 86% lower costs for their vision AI workload.
Amazon Polly Generative Engine Receives Upgrade, Adds 14 New Voices and Hindi/Finnish Support
Amazon Polly's Generative engine has been upgraded to provide more natural and accurate speech output with low latency. This update includes 14 new voices across 9 locales, with general availability for Hindi and Finnish voices for the first time, enhancing its utility for various applications requiring scalable voice output.
IskayPet Implements Amazon Bedrock Managed Knowledge Bases for Corporate Knowledge Unification
IskayPet, a pet care company, deployed a corporate knowledge assistant using Amazon Bedrock Managed Knowledge Bases to centralize operational information across its multi-country, multi-language operations. This solution addresses the challenge of employees manually searching hundreds of SharePoint documents for answers to everyday questions, improving efficiency and accuracy.
TwelveLabs Pegasus 1.5 Video Language Model Now Available on Amazon Bedrock
TwelveLabs' Pegasus 1.5, a video language model, is now available on Amazon Bedrock. This update allows for the analysis of long-form video to generate structured, timestamped metadata, enabling better searchability and utilization of video content.
Deepgram optimizes Nova-3 voice AI model for AWS Trainium2, improving efficiency
Deepgram collaborated with AWS to port its Nova-3 speech-to-text model to AWS Trainium2, achieving 11-70% more audio processing per dollar for batch workloads. This optimization allows Deepgram to scale its voice AI services more efficiently and potentially offer lower prices to customers.
Amazon SageMaker HyperPod integrates new Ray capabilities for distributed ML workloads
Amazon SageMaker HyperPod now includes new capabilities for Ray, an open-source framework for scaling distributed Python workloads. This integration simplifies the management of Ray clusters on HyperPod's infrastructure, providing built-in fault tolerance and improved observability for large-scale machine learning.
Claude Sonnet 5.5 now available on Amazon Bedrock and Claude Platform on AWS
Claude Sonnet 5.5, an updated AI model, is now available on Amazon Bedrock and Claude Platform on AWS. This version offers improved efficiency for coding and knowledge work at a lower cost and faster speed, making it suitable for continuous or large-scale AI workloads within AWS infrastructure.