Some sources failed to update13 / 14 sources available · Checked every 10 minutes · 30-day retention · 168 deduplicated items · Updated Sep 29, 2026, 3:00 PM Beijing time
Model releaseOfficialAWS Machine LearningAI summary

Grok 4.7 is now available on Amazon Bedrock

xAI's Grok 4.7 is now available on Amazon Bedrock as a frontier model designed for coding, long-running agents, and knowledge work. The model features a 500K token context window and four configurable reasoning effort levels. It can be accessed through the Responses, Chat Completions, and Converse APIs.

SOURCEGrok 4.7 is now available on Amazon Bedrock

Read source
Model releaseCommunitySimon Willison2 sourcesAI summary

Introducing Claude Sonnet 5.5 on AWS

Anthropic's Claude Sonnet 5.5 is now generally available on Amazon Bedrock and Claude Platform on AWS. The updated Sonnet model delivers improved efficiency for focused coding and knowledge work tasks, offering faster processing speeds and a lower cost per task compared to previous versions.

IndustryResearchMicrosoft ResearchAI summary

One year in: How Microsoft Research Asia – Singapore is advancing research, partnership and talent for real-world impact

Microsoft Research Asia – Singapore marked its first anniversary, having established a strong foundation for frontier AI research. The lab has deepened collaboration across government, academia, and industry sectors in the Asia-Pacific region, focusing on translating research into real-world impact through partnerships and talent development.

SOURCEOne year in: How Microsoft Research Asia – Singapore is advancing research, partnership and talent for real-world impact

Read source
Safety and policyOfficialOpenAIAI summary

How we will do better for Australia

OpenAI issued an apology for incidents involving Australian government websites and outlined stronger safeguards to strengthen Australia's cyber defenses. The company committed to enhanced security measures and support to prevent future occurrences.

IndustryOfficialMistral AIAI summary

Hallo, Deutschland!

Mistral AI announced the opening of a new research hub in Munich, Germany, dedicated to Physics AI and Industrial AI research. The Munich hub will partner with German industry to advance applied AI in these specialized domains, strengthening Mistral's presence in Europe.

Tools and tutorialsOfficialHugging FaceAI summary

Holo4: powering generalist computer-use agents

Hugging Face releases Holo4, a system designed to power generalist computer-use agents. The project enables AI systems to interact with and control computer interfaces for diverse tasks. Details about architecture, capabilities, and availability are available on the Hugging Face platform.

SOURCEHolo4: powering generalist computer-use agents

Read source
IndustryOfficialOpenAIAI summary

The Lenfest Institute grows landmark program with expanded OpenAI support

OpenAI is expanding the Lenfest AI Collaborative and Fellowship Program with $5 million in funding and up to $5 million in software credits and engineering support. The program supports local news organizations in developing AI applications for journalism. This expanded commitment aims to help newsrooms responsibly integrate AI tools while maintaining editorial independence and journalistic integrity.

SOURCEThe Lenfest Institute grows landmark program with expanded OpenAI support

Read source
ResearchResearchApple Machine LearningAI summary

Faster Rates for Federated Variational Inequalities

This paper addresses convergence rates in federated optimization for solving stochastic variational inequalities (VIs). The authors identify a significant gap between existing convergence rates and state-of-the-art bounds in federated convex optimization, proposing improved algorithmic approaches. The research contributes theoretical advances to federated learning methods for non-cooperative game settings.

SOURCEFaster Rates for Federated Variational Inequalities

Read source
Safety and policyResearchMETRAI summary

Implementing and Evaluating a Basic Per-Action Monitor for Safer Evals

METR developed and deployed a basic live per-action monitor to enhance safety during AI agent evaluations. Inspired by recent incidents from OpenAI, Anthropic, and UK AISI, the monitor detects actions that could plausibly cause real-world harm. The system aims to reliably flag potentially dangerous agent behaviors during evaluation runs.

SOURCEImplementing and Evaluating a Basic Per-Action Monitor for Safer Evals

Read source
Tools and tutorialsOfficialAWS Machine LearningAI summary

NarrateAI: production-ready LLM quality assurance on Amazon Bedrock

NarrateAI provides production-ready LLM quality assurance on Amazon Bedrock through five key techniques: adaptive pipeline orchestration, cross-account multi-model failover, real-time streaming evaluation, composite evaluation, and data accuracy verification. These methods achieve approximately 99% numerical accuracy while maintaining real-time response streaming capabilities.

SOURCENarrateAI: production-ready LLM quality assurance on Amazon Bedrock

Read source
Model releaseOfficialGoogle DeepMindAI summary

Introducing Gemini 3.8 Live with Live Avatar

Google DeepMind introduced Gemini 3.8 Live with Live Avatar functionality. This release brings interactive avatar capabilities to Gemini 3.8, enabling more dynamic and visually engaging AI interactions. The update represents Google DeepMind's latest advancement in multimodal AI technology.

SOURCEIntroducing Gemini 3.8 Live with Live Avatar

Read source
ResearchOfficialNVIDIAAI summary

How Open Science Can Help Researchers Prepare for the Next Pandemic

NVIDIA has joined a coalition of global research organizations working to apply open science methodologies for pandemic preparedness. The initiative aims to ensure that when the next pandemic emerges, researchers can respond more rapidly than they did with COVID-19, which benefited from decades of prior coronavirus research that enabled record vaccine development.

SOURCEHow Open Science Can Help Researchers Prepare for the Next Pandemic

Read source