Today, AWS announces increased Amazon Elastic Block Store (Amazon EBS) performance for Amazon EC2 C8gn, M8gn, and R8gn instances in 48xlarge and metal-48xl sizes.

EC2 C8gn, M8gn, and R8gn instances are network optimized instances powered by AWS Graviton4 processors and latest 6th generation AWS Nitro Cards. With the latest enhancements to AWS Nitro System, we have doubled the maximum EBS performance on these instances in 48xlarge and metal-48xl sizes, from 60 Gbps of EBS bandwidth and 240,000 IOPS to 120 Gbps of EBS bandwidth and 480,000 IOPS. Customers running network-intensive workloads while requiring additional block storage performance such as data analytics and high-performance file systems can benefit from the improved EBS performance.

All existing and new C8gn, M8gn, and R8gn instances in 48xlarge and metal-48xl sizes launched starting today will benefit from this performance increase at no additional cost. For running instances, customers can stop and start instances to enable this performance increase. The higher EBS performance is available in all AWS regions where these instance types are generally available today.

To learn more, see Amazon C8gn, M8gn, and R8gn Instances and EBS-optimized instance types. 

​Today, AWS announces increased Amazon Elastic Block Store (Amazon EBS) performance for Amazon EC2 C8gn, M8gn, and R8gn instances in 48xlarge and metal-48xl sizes.
EC2 C8gn, M8gn, and R8gn instances are network optimized instances powered by AWS Graviton4 processors and latest 6th generation AWS Nitro Cards. With the latest enhancements to AWS Nitro System, we have doubled the maximum EBS performance on these instances in 48xlarge and metal-48xl sizes, from 60 Gbps of EBS bandwidth and 240,000 IOPS to 120 Gbps of EBS bandwidth and 480,000 IOPS. Customers running network-intensive workloads while requiring additional block storage performance such as data analytics and high-performance file systems can benefit from the improved EBS performance.
All existing and new C8gn, M8gn, and R8gn instances in 48xlarge and metal-48xl sizes launched starting today will benefit from this performance increase at no additional cost. For running instances, customers can stop and start instances to enable this performance increase. The higher EBS performance is available in all AWS regions where these instance types are generally available today.
To learn more, see Amazon C8gn, M8gn, and R8gn Instances and EBS-optimized instance types.   

• Categoría: Noticias de Microsoft

abril 14, 2026

MAI-Image-2-Efficient: calidad insignia, costo 41% menor

Por: Equipo de MAI Superintelligence.

Disponible ahora en Microsoft Foundryy MAI Playground

Construimos MAI-Image-2 para que fuera nuestro mejor modelo de texto a imagen — fotorrealista, expresivo, con texto dentro de la imagen fiable.

Hoy hacemos todo eso más rápido y barato.

Conozcan MAI-Image-2-Efficient.

Calidad lista para producción. Construido para la velocidad y la escala. 22% más rápido y 4 veces más eficiente¹. Y con un precio casi un 41% más bajo — 5 dólares por cada 1 millón de tokens de entrada de texto, 19,50 dólares por cada 1 millón de tokens de salida de imagen.

Eso no es solo más rápido que nuestro propio modelo insignia. Es un 40% más rápido en promedio que otros modelos principales de texto a imagen².

Dos modelos, dos trabajos

MAI-Image-2-Efficient es su caballo de batalla de producción. Úsenlo cuando necesiten volumen, rapidez y un control estricto de costes — fotos de producto, creatividades de marketing, maquetas de interfaz, activos de marca, pipelines por lotes. Este modelo gestiona textos cortos como encabezados y etiquetas de forma limpia, y está diseñado para funcionar en flujos de trabajo interactivos en tiempo real sin despeinarse.

MAI-Image-2 es su herramienta de precisión. Úsenlo cuando el encargo requiera la mayor fidelidad: retratos, escenas fotorrealistas, looks estilizados como anime o ilustración, y texto más largo o complejo en imagen. Este es el modelo para los entregables finales donde cada detalle importa.

Empiecen a construir ahora

MAI-Image-2-Efficient está disponible hoy en Microsoft Foundry y MAI Playground³. Sin lista de espera, sin vista previa — sólo actívenlo y listo. También se va a desplegar en Copilot y Bing, con más superficies como PowerPoint más adelante.

Socios como Shutterstock ya lo han probado con resultados prometedores:

«MAI-Image-2-Efficient muestra un fuerte progreso en fidelidad rápida y usabilidad creativa a lo largo de una variedad de flujos de trabajo. En nuestro trabajo de evaluación, analizamos con detenimiento cómo los modelos traducen la intención en resultados consistentes y listos para producción, y este modelo va en la dirección correcta. Ese nivel de fiabilidad es lo que en verdad importa cuando los equipos pasan de la experimentación al uso en el mundo real.» – Vanessa Salvo, directora de producto principal, Shutterstock

Esto es solo el principio. Más modelos llegarán más adelante— estén atentos.

Descargar tarjeta de modelo

1. Tal como se probó el 13 de abril de 2026. Comparado con MAI-Image-2 cuando se normaliza por latencia y uso de GPU. Rendimiento por GPU vs MAI-Image-2 en NVIDIA H100 a 1024×1024; se midió con tamaños de lote optimizados y objetivos de latencia coincididos. Los resultados varían según el tamaño del lote, la concurrencia y las restricciones de latencia.
2. Tal como se probó el 13 de abril de 2026. Comparado con Gemini 3.1 Flash (alto razonamiento), Gemini 3.1 Flash Image y Gemini 3 Pro Image: medido a p50 de latencia mediante AI Studio API (1:1, 1K imágenes; razonamiento mínimo salvo que se indique; búsqueda web deshabilitada). MAI-Image-2, MAI-Image-2e, GPT-Image-1.5-High: Medido a una latencia p50 mediante Foundry API.
3. MAI Playground está disponible en mercados seleccionados, incluidos Estados Unidos. Más adelante en los países de la UE.

Etiquetas:

• IA
• Innovación

The post MAI-Image-2-Efficient: calidad insignia, costo 41% menor appeared first on Source LATAM.

​The post MAI-Image-2-Efficient: calidad insignia, costo 41% menor appeared first on Source LATAM.  

NVIDIA’s Nemotron-3-Super-120B, Qwen3.5-9B, and Qwen3.5-27B models are now available on Amazon SageMaker JumpStart, expanding the portfolio of foundation models available to AWS customers. These three models bring specialized capabilities spanning agentic reasoning, multilingual coding, and advanced instruction following, enabling customers to deploy high-performance, scalable AI solutions on AWS infrastructure.

These models address different enterprise AI challenges with specialized capabilities:
Nemotron-3-Super-120B is optimized for collaborative agents and high-volume workloads such as IT ticket automation. It employs a hybrid Latent Mixture-of-Experts (LatentMoE) architecture with Mamba-2 and MoE layers, enabling strong agentic, reasoning, and conversational capabilities useful for multi-agent applications like software development and cybersecurity triaging.
Qwen 3.5 9B excels in multilingual coding, instruction following, and long-horizon planning, automating software development workflows and executing complex, multi-step office tasks. Its compact design balances efficiency and performance for resource-constrained environments.
Qwen 3.5 27B provides deeper contextual understanding, extended reasoning capabilities, and enhanced spatial/complex scenario comprehension, ideal for advanced multimodal reasoning and large-scale document processing.
With SageMaker JumpStart, customers can deploy any of these models with just a few clicks to address their specific AI use cases.

To get started with these models, navigate to the SageMaker JumpStart model catalog in the SageMaker console or use the SageMaker Python SDK to deploy the models to your AWS account. For more information about deploying and using foundation models in SageMaker JumpStart, see the Amazon SageMaker JumpStart documentation.

​NVIDIA’s Nemotron-3-Super-120B, Qwen3.5-9B, and Qwen3.5-27B models are now available on Amazon SageMaker JumpStart, expanding the portfolio of foundation models available to AWS customers. These three models bring specialized capabilities spanning agentic reasoning, multilingual coding, and advanced instruction following, enabling customers to deploy high-performance, scalable AI solutions on AWS infrastructure. These models address different enterprise AI challenges with specialized capabilities: Nemotron-3-Super-120B is optimized for collaborative agents and high-volume workloads such as IT ticket automation. It employs a hybrid Latent Mixture-of-Experts (LatentMoE) architecture with Mamba-2 and MoE layers, enabling strong agentic, reasoning, and conversational capabilities useful for multi-agent applications like software development and cybersecurity triaging. Qwen 3.5 9B excels in multilingual coding, instruction following, and long-horizon planning, automating software development workflows and executing complex, multi-step office tasks. Its compact design balances efficiency and performance for resource-constrained environments. Qwen 3.5 27B provides deeper contextual understanding, extended reasoning capabilities, and enhanced spatial/complex scenario comprehension, ideal for advanced multimodal reasoning and large-scale document processing. With SageMaker JumpStart, customers can deploy any of these models with just a few clicks to address their specific AI use cases. To get started with these models, navigate to the SageMaker JumpStart model catalog in the SageMaker console or use the SageMaker Python SDK to deploy the models to your AWS account. For more information about deploying and using foundation models in SageMaker JumpStart, see the Amazon SageMaker JumpStart documentation.  

Amazon Quick now supports document-level access controls (ACLs) for Google Drive knowledge bases, enabling organizations to maintain native Google Drive permissions when indexing content. Quick combines ACL replication for efficient pre-retrieval filtering with an additional layer of real-time permission checks directly with Google Drive at query time. This dual approach means you get the performance benefits of indexed ACLs while also guarding against stale or incorrectly mapped permission data. When a user submits a query, Quick verifies their current permissions with Google Drive before generating a response—ensuring answers are based on live access rights.

With document-level access controls, Amazon Quick now respects individual file and folder permissions from Google Drive. This feature is available in all AWS Regions where Amazon Quick is available.

To get started, create or update a Google Drive knowledge base in the Amazon Quick console and configure document-level access controls in your integration settings. For more information, see Google Drive integration in the Amazon Quick User Guide.

​Amazon Quick now supports document-level access controls (ACLs) for Google Drive knowledge bases, enabling organizations to maintain native Google Drive permissions when indexing content. Quick combines ACL replication for efficient pre-retrieval filtering with an additional layer of real-time permission checks directly with Google Drive at query time. This dual approach means you get the performance benefits of indexed ACLs while also guarding against stale or incorrectly mapped permission data. When a user submits a query, Quick verifies their current permissions with Google Drive before generating a response—ensuring answers are based on live access rights. With document-level access controls, Amazon Quick now respects individual file and folder permissions from Google Drive. This feature is available in all AWS Regions where Amazon Quick is available.
To get started, create or update a Google Drive knowledge base in the Amazon Quick console and configure document-level access controls in your integration settings. For more information, see Google Drive integration in the Amazon Quick User Guide.  

Amazon Redshift further optimizes the processing of top-k queries (queries with ORDER BY and LIMIT clauses) by intelligently skipping irrelevant data blocks to return results faster, dramatically reducing the amount of data processed. This optimization reorders and efficiently adjusts the data blocks to be read based on the ORDER BY column’s min/max values, maintaining only the K most qualifying rows in memory. When the ORDER BY column is sorted or partially sorted, Amazon Redshift now processes only the minimal data blocks needed rather than scanning entire tables, eliminating unnecessary I/O and compute overhead.

This enhancement particularly benefits top-k queries when the data permanently stores in descending order (ORDER BY … DESC LIMIT K) on large tables where qualifying rows are appended at the end of the data storage. Common examples include:

• Finding the k most recent orders from millions or billions of transactions
• Retrieving top-k best performing products or k worst performing products (top-k in descending order) from your sales catalog containing hundreds of thousands stock keeping units (SKUs) and millions or billions of sales transactions associated with all product SKUs in your sales catalog
• Finding the top-k most recent or top-k oldest (top k in descending order) prompts inferred by a foundational large language model (LLM) out of billions of prompts.

With this new optimization, top-k query performance improves dramatically. This optimization for top-k queries is now available in Amazon Redshift at no additional cost starting with patch release P199 across all AWS regions where Amazon Redshift is available. This optimization automatically applies to eligible queries without requiring any query rewrites or configuration changes.

​Amazon Redshift further optimizes the processing of top-k queries (queries with ORDER BY and LIMIT clauses) by intelligently skipping irrelevant data blocks to return results faster, dramatically reducing the amount of data processed. This optimization reorders and efficiently adjusts the data blocks to be read based on the ORDER BY column’s min/max values, maintaining only the K most qualifying rows in memory. When the ORDER BY column is sorted or partially sorted, Amazon Redshift now processes only the minimal data blocks needed rather than scanning entire tables, eliminating unnecessary I/O and compute overhead.
This enhancement particularly benefits top-k queries when the data permanently stores in descending order (ORDER BY … DESC LIMIT K) on large tables where qualifying rows are appended at the end of the data storage. Common examples include:

Finding the k most recent orders from millions or billions of transactions
Retrieving top-k best performing products or k worst performing products (top-k in descending order) from your sales catalog containing hundreds of thousands stock keeping units (SKUs) and millions or billions of sales transactions associated with all product SKUs in your sales catalog
Finding the top-k most recent or top-k oldest (top k in descending order) prompts inferred by a foundational large language model (LLM) out of billions of prompts.

With this new optimization, top-k query performance improves dramatically. This optimization for top-k queries is now available in Amazon Redshift at no additional cost starting with patch release P199 across all AWS regions where Amazon Redshift is available. This optimization automatically applies to eligible queries without requiring any query rewrites or configuration changes.  

Amazon CloudWatch Logs Insights saved queries now support parameters, allowing you to pass values to reusable query templates with placeholders. This eliminates the need to maintain multiple copies of nearly identical queries that differ only in specific values such as log levels, service names, or time intervals.

You can define up to 20 parameters in a query, with each parameter supporting optional default values. For example, you can create a single template to query logs by severity level (such as ERROR or WARN) and pass different service names each time you run it. To execute a query with parameters, invoke it using the query name prefixed with $ and pass your parameter values, such as $ErrorsByService(logLevel=»ERROR», serviceName=»OrderEntry»). You can also use multiple saved queries with parameters together for complex log analysis, significantly reducing query maintenance overhead while improving reusability.

Saved queries with parameters are available in all commercial AWS regions. You can create and use saved queries with parameters using the Amazon CloudWatch console, AWS Command Line Interface (AWS CLI), AWS Cloud Development Kit (AWS CDK), and AWS SDKs. To learn more, see the Amazon CloudWatch Logs documentation.

​Amazon CloudWatch Logs Insights saved queries now support parameters, allowing you to pass values to reusable query templates with placeholders. This eliminates the need to maintain multiple copies of nearly identical queries that differ only in specific values such as log levels, service names, or time intervals. You can define up to 20 parameters in a query, with each parameter supporting optional default values. For example, you can create a single template to query logs by severity level (such as ERROR or WARN) and pass different service names each time you run it. To execute a query with parameters, invoke it using the query name prefixed with $ and pass your parameter values, such as $ErrorsByService(logLevel=»ERROR», serviceName=»OrderEntry»). You can also use multiple saved queries with parameters together for complex log analysis, significantly reducing query maintenance overhead while improving reusability. Saved queries with parameters are available in all commercial AWS regions. You can create and use saved queries with parameters using the Amazon CloudWatch console, AWS Command Line Interface (AWS CLI), AWS Cloud Development Kit (AWS CDK), and AWS SDKs. To learn more, see the Amazon CloudWatch Logs documentation.  

Amazon OpenSearch Serverless introduces support for Derived Source, a new feature that can help reduce the amount of storage required for your OpenSearch Service collections. With derived source support, you can skip storing source fields and dynamically derive them when required. 

With Derived Source, OpenSearch Serverless reconstructs the _source field on the fly using the values already stored in the index, eliminating the need to maintain a separate copy of the original document. This can significantly reduce storage consumption, particularly for time-series and log analytics collections where documents contain many indexed fields. You can enable derived source at the index level when creating or updating index mappings.

Derived Source support is available today in all AWS Regions where Amazon OpenSearch Serverless is supported. For more information, see the Amazon OpenSearch Serverless documentation.

​Amazon OpenSearch Serverless introduces support for Derived Source, a new feature that can help reduce the amount of storage required for your OpenSearch Service collections. With derived source support, you can skip storing source fields and dynamically derive them when required. 
With Derived Source, OpenSearch Serverless reconstructs the _source field on the fly using the values already stored in the index, eliminating the need to maintain a separate copy of the original document. This can significantly reduce storage consumption, particularly for time-series and log analytics collections where documents contain many indexed fields. You can enable derived source at the index level when creating or updating index mappings.
Derived Source support is available today in all AWS Regions where Amazon OpenSearch Serverless is supported. For more information, see the Amazon OpenSearch Serverless documentation.  

Today we are announcing the release of the Aurora DSQL Connector for PHP (PDO_PGSQL) that makes it easy to build PHP applications on Aurora DSQL. The PHP Connector streamlines authentication and eliminates security risks associated with traditional user-generated passwords by automatically generating tokens for each connection, ensuring valid tokens are always used while maintaining full compatibility with existing PDO_PGSQL features.

The connector handles IAM token generation, SSL configuration, and connection pooling, enabling customers to scale from simple scripts to production workloads without changing their authentication approach. It also provides opt-in optimistic concurrency control (OCC) retry with exponential backoff, custom IAM credential providers, and AWS profile support, making it easier to develop client retry logic and manage AWS credentials.

To get started, visit the Connectors for Aurora DSQL documentation page. For code examples, visit our GitHub page for the PHP connector. Get started with Aurora DSQL for free with the AWS Free Tier. To learn more about Aurora DSQL, visit the webpage.    

​Today we are announcing the release of the Aurora DSQL Connector for PHP (PDO_PGSQL) that makes it easy to build PHP applications on Aurora DSQL. The PHP Connector streamlines authentication and eliminates security risks associated with traditional user-generated passwords by automatically generating tokens for each connection, ensuring valid tokens are always used while maintaining full compatibility with existing PDO_PGSQL features. The connector handles IAM token generation, SSL configuration, and connection pooling, enabling customers to scale from simple scripts to production workloads without changing their authentication approach. It also provides opt-in optimistic concurrency control (OCC) retry with exponential backoff, custom IAM credential providers, and AWS profile support, making it easier to develop client retry logic and manage AWS credentials. To get started, visit the Connectors for Aurora DSQL documentation page. For code examples, visit our GitHub page for the PHP connector. Get started with Aurora DSQL for free with the AWS Free Tier. To learn more about Aurora DSQL, visit the webpage.      

Starting today, Amazon Elastic Compute Cloud (Amazon EC2) R8i and R8i-flex instances are available in the AWS GovCloud (US-West) Region. These instances are powered by custom Intel Xeon 6 processors, available only on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud. The R8i and R8i-flex instances offer up to 15% better price-performance, and 2.5x more memory bandwidth compared to previous generation Intel-based instances. They deliver 20% higher performance than R7i instances, with even higher gains for specific workloads. They are up to 30% faster for PostgreSQL databases, up to 60% faster for NGINX web applications, and up to 40% faster for AI deep learning recommendation models compared to R7i.

R8i-flex, our first memory-optimized Flex instances, are the easiest way to get price performance benefits for a majority of memory-intensive workloads. They offer the most common sizes, from large to 16xlarge, and are a great first choice for applications that don’t fully utilize all compute resources.

R8i instances are a great choice for all memory-intensive workloads, especially for workloads that need the largest instance sizes or continuous high CPU usage. R8i instances offer 13 sizes including 2 bare metal sizes and the new 96xlarge size for the largest applications. R8i instances are SAP-certified and deliver 142,100 aSAPS, delivering exceptional performance for mission-critical SAP workloads.

To get started, sign in to the AWS Management Console. For more information about the R8i and R8i-flex instances visit the AWS News blog.

​Starting today, Amazon Elastic Compute Cloud (Amazon EC2) R8i and R8i-flex instances are available in the AWS GovCloud (US-West) Region. These instances are powered by custom Intel Xeon 6 processors, available only on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud. The R8i and R8i-flex instances offer up to 15% better price-performance, and 2.5x more memory bandwidth compared to previous generation Intel-based instances. They deliver 20% higher performance than R7i instances, with even higher gains for specific workloads. They are up to 30% faster for PostgreSQL databases, up to 60% faster for NGINX web applications, and up to 40% faster for AI deep learning recommendation models compared to R7i. R8i-flex, our first memory-optimized Flex instances, are the easiest way to get price performance benefits for a majority of memory-intensive workloads. They offer the most common sizes, from large to 16xlarge, and are a great first choice for applications that don’t fully utilize all compute resources. R8i instances are a great choice for all memory-intensive workloads, especially for workloads that need the largest instance sizes or continuous high CPU usage. R8i instances offer 13 sizes including 2 bare metal sizes and the new 96xlarge size for the largest applications. R8i instances are SAP-certified and deliver 142,100 aSAPS, delivering exceptional performance for mission-critical SAP workloads. To get started, sign in to the AWS Management Console. For more information about the R8i and R8i-flex instances visit the AWS News blog.  

Starting today, Amazon EC2 M8i and M8i-flex instances are now available in AWS GovCloud (US-West) Region. These instances are powered by custom Intel Xeon 6 processors, available only on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud. The M8i and M8i-flex instances offer up to 15% better price-performance, and 2.5x more memory bandwidth compared to previous generation Intel-based instances. They deliver up to 20% better performance than M7i and M7i-flex instances, with even higher gains for specific workloads. The M8i and M8i-flex instances are up to 30% faster for PostgreSQL databases, up to 60% faster for NGINX web applications, and up to 40% faster for AI deep learning recommendation models compared to M7i and M7i-flex instances.

M8i-flex are the easiest way to get price performance benefits for a majority of general-purpose workloads like web and application servers, microservices, small and medium data stores, virtual desktops, and enterprise applications. They offer the most common sizes, from large to 16xlarge, and are a great first choice for applications that don’t fully utilize all compute resources.

M8i instances are a great choice for all general purpose workloads, especially for workloads that need the largest instance sizes or continuous high CPU usage. The SAP-certified M8i instances offer 13 sizes including 2 bare metal sizes and the new 96xlarge size for the largest applications.

To get started, sign in to the AWS Management Console. For more information about the new instances, visit the M8i and M8i-flex instance page or visit the AWS News blog.

​Starting today, Amazon EC2 M8i and M8i-flex instances are now available in AWS GovCloud (US-West) Region. These instances are powered by custom Intel Xeon 6 processors, available only on AWS, delivering the highest performance and fastest memory bandwidth among comparable Intel processors in the cloud. The M8i and M8i-flex instances offer up to 15% better price-performance, and 2.5x more memory bandwidth compared to previous generation Intel-based instances. They deliver up to 20% better performance than M7i and M7i-flex instances, with even higher gains for specific workloads. The M8i and M8i-flex instances are up to 30% faster for PostgreSQL databases, up to 60% faster for NGINX web applications, and up to 40% faster for AI deep learning recommendation models compared to M7i and M7i-flex instances. M8i-flex are the easiest way to get price performance benefits for a majority of general-purpose workloads like web and application servers, microservices, small and medium data stores, virtual desktops, and enterprise applications. They offer the most common sizes, from large to 16xlarge, and are a great first choice for applications that don’t fully utilize all compute resources. M8i instances are a great choice for all general purpose workloads, especially for workloads that need the largest instance sizes or continuous high CPU usage. The SAP-certified M8i instances offer 13 sizes including 2 bare metal sizes and the new 96xlarge size for the largest applications. To get started, sign in to the AWS Management Console. For more information about the new instances, visit the M8i and M8i-flex instance page or visit the AWS News blog.