Gen AI Live

Anthropic highlights that Claude Code is built with security as a core foundation, supported by compliance standards like SOC 2 and ISO 27001.  The platform includes safeguards such as sandboxing, permission controls, and secure deployment options to protect code and data across environments.  

It follows principles like isolation, least privilege, and defense in depth to reduce risks when running AI agents.  The system also addresses threats like prompt injection and unintended agent actions by combining model level protections with infrastructure controls.

Overall, the approach focuses on enabling safe, scalable use of AI coding agents in both individual and enterprise workflows.

The U.S. Pentagon has signed agreements with seven leading AI companies, including OpenAI, Google, Microsoft, Amazon Web Services, NVIDIA, SpaceX, and Reflection, to deploy their technologies on classified defense networks.  

These systems will support tasks such as data synthesis, situational awareness, and decision making in complex military operations.  The initiative aims to transform the military into an AI driven force, improving speed and accuracy across operations.  

Notably, Anthropic was excluded from the agreements due to disputes over usage safeguards, highlighting tensions between safety concerns and defense requirements in AI deployment.

Grok 4.3 delivers competitive benchmark performance with a score of 53 on the Artificial Analysis Intelligence Index, placing it well above the average of similar models.  It shows strong gains in reasoning, coding, and agent tasks, with notable improvements in benchmarks like GDPval and instruction following.  

The model also stands out for speed, reaching over 100 tokens per second, and offers significantly lower pricing compared to competitors.  

However, it still trails leading models such as GPT 5.5 and Claude Opus 4.7 in overall intelligence rankings, highlighting a tradeoff between performance and cost efficiency.

‍

NVIDIA GeForce NOW is a cloud gaming platform that lets users stream PC games directly from powerful remote servers instead of running them locally. It delivers high performance gameplay with features like ray tracing, high frame rates, and low latency across devices such as laptops, smartphones, and smart TVs.

Users can connect their existing game libraries from platforms like Steam or Epic Games and play instantly without downloads or updates.  

The service offers multiple subscription tiers, including a free option and premium plans with higher performance, making advanced gaming more accessible and scalable for a global audience.

Genspark has announced a global partnership with Microsoft to embed its AI agents directly داخل Microsoft 365 applications and Agent 365. This integration allows users to create presentations, analyze data, and draft documents within tools like Word, Excel, and PowerPoint without switching platforms.  

The collaboration focuses on reducing workflow friction by bringing AI into the tools professionals already use daily. Built on Azure, the system ensures enterprise scale, security, and performance.  

By combining Genspark’s agentic AI with Microsoft’s ecosystem, the partnership aims to transform AI from a separate tool into an embedded layer that drives productivity and automation across organizations.

‍

OpenAI is accelerating its Stargate project to build the compute infrastructure required for the Intelligence Age, focusing on scaling data centers and securing massive computing capacity. The company has already reached its 10 gigawatt target for AI compute in the US years ahead of schedule, with rapid additions in recent months.  

This infrastructure supports training and deploying more advanced AI systems while lowering costs and improving reliability.

OpenAI is working with partners, governments, and local communities to expand this ecosystem, aiming to make AI accessible for businesses, developers, and individuals at scale.

OpenAI has introduced Advanced Account Security, a new opt-in feature designed to protect ChatGPT accounts from unauthorized access, especially for users handling sensitive information. It replaces traditional passwords with passkeys or hardware security keys, making accounts resistant to phishing and credential theft.  

Once enabled, email and SMS recovery options are disabled, and users must rely on secure credentials and recovery keys.  

The system also shortens login sessions, provides alerts for new sign-ins, and allows users to monitor active sessions.  This reflects a shift toward stronger, zero trust security models for protecting AI accounts.

OpenAI explained the origin of unusual “goblin” and “gremlin” references in GPT 5.5, especially within its coding agent Codex. The behavior likely emerged from how large language models generate patterns and metaphors, sometimes amplified by agent tools like OpenClaw that add context and instructions.  

To address this, OpenAI added explicit instructions in the system prompt to prevent such references unless directly relevant.  

The issue highlights how advanced models can develop unexpected quirks and why alignment, prompt design, and behavioral safeguards are critical for maintaining reliable, professional outputs in real world applications.

‍

Google has introduced a major update to Gemini that allows users to create fully formatted, downloadable files directly within the chat interface. Users can generate Google Docs, Sheets, Slides, PDFs, Word, Excel, and more using a single prompt, eliminating the need for copying, pasting, and manual formatting across applications.  

This feature bridges the gap between brainstorming and execution by turning ideas into ready-to-share outputs instantly. It supports multiple formats, including CSV, LaTeX, Markdown, and text files, making it useful for both business and technical workflows.  

The update improves productivity by keeping the entire workflow inside Gemini.

‍

NVIDIA has launched Nemotron 3 Nano Omni, a new multimodal AI model designed to act as the core engine for agentic AI systems. It integrates text, vision, and speech capabilities, enabling more advanced reasoning and real world task execution.

The model focuses on efficiency, delivering high accuracy while reducing compute costs, making it suitable for enterprise deployment at scale.  

It is part of the broader Nemotron 3 family, which provides a full stack of models and tools for building production ready AI agents across use cases like automation, coding, and multimodal workflows.

OpenAI is strengthening its collaboration with Amazon Web Services to support the growing demand for large scale AI infrastructure and enterprise deployment.

Its models, including advanced reasoning systems and tools like Codex, are now available through AWS platforms such as Bedrock, allowing organizations to build, deploy, and manage AI applications more easily.  

The partnership also focuses on developing stateful runtime environments that enable long running, context aware AI agents integrated with enterprise systems.  This move reflects OpenAI’s shift toward a multi cloud strategy, improving scalability, flexibility, and access to compute needed for next generation AI systems.

Anthropic introduced Claude for creative work, positioning its AI as an active collaborator inside professional creative tools. The update adds connectors for platforms like Adobe Creative Cloud, Blender, Autodesk Fusion, and Ableton, allowing Claude to directly interact with software environments and automate repetitive tasks.  

Instead of acting as a standalone assistant, Claude now works within existing workflows, helping with tasks like editing, scripting, and production support.

The goal is to accelerate ideation, expand creative capabilities, and enable users to handle larger, more complex projects while keeping human creativity at the center of the process.

‍

OpenAI and Microsoft have significantly renegotiated their partnership, marking a shift from a tightly coupled alliance to a more flexible, multi cloud model. The updated agreement removes the long standing AGI clause, which previously governed access and revenue terms tied to advanced AI breakthroughs.

OpenAI can now offer its models across multiple cloud providers, expanding enterprise reach and reducing dependency on Microsoft.  Microsoft remains the primary cloud partner and retains non exclusive rights to OpenAI technology through 2032.  

The deal also simplifies financial terms, introducing capped payments and clearer timelines, reflecting the growing scale and competition in the AI industry.

‍

OpenAI and Microsoft have moved into the next phase of their partnership through a new agreement designed to strengthen long term collaboration and support evolving AI needs. The partnership continues to focus on building advanced AI systems, scaling infrastructure, and delivering tools to businesses and users worldwide.

Microsoft remains a key partner in cloud, research, and product integration, while both companies aim to expand innovation and ensure responsible AI deployment. The agreement also supports OpenAI’s structural evolution and future growth plans.

Overall, this phase reflects a balance between deep collaboration and the flexibility needed to scale AI globally.

‍

Anthropic is experimenting with a new BugCrawl tool designed to enhance bug detection in Claude Code by using multiple AI agents to scan and analyze code. The system builds on earlier features like Code Review, where parallel agents examine pull requests, detect logic errors, and prioritize vulnerabilities by severity.  

BugCrawl aims to further automate this process by continuously crawling codebases and identifying issues before deployment.

This reflects a broader shift toward AI driven software validation, where automated systems handle large volumes of generated code, reduce human review effort, and improve reliability in fast moving development environments.

OpenAI has outlined a set of principles that guide how it builds and deploys advanced AI systems. These principles focus on ensuring that AI benefits all of humanity, maintaining long term safety, and leading in technical innovation.

The company emphasizes responsible development, including reducing risks from powerful systems while enabling broad access and impact. It also highlights the importance of collaboration, governance, and adapting to evolving challenges as AI capabilities grow.  

The framework reflects a balance between rapid progress and careful oversight, ensuring that AI systems are developed in ways that align with societal needs and long term global benefit.

‍

Anthropic has announced a strategic partnership with NEC to scale enterprise AI adoption in Japan by building one of the country’s largest AI native engineering workforces. As part of the collaboration, NEC will deploy Anthropic’s Claude AI tools to around 30,000 employees and become its first Japan based global partner.  

The companies will jointly develop secure, industry specific AI solutions tailored for sectors such as finance, manufacturing, and local government.  

This initiative focuses on integrating AI into real business workflows while leveraging NEC’s domain expertise and Anthropic’s advanced models to drive productivity, automation, and innovation at scale.

‍

DeepSeek introduced V4 Preview as an open source release with a 1M token context window, setting a new standard for long context efficiency. The lineup includes DeepSeek V4 Pro, a high performance model with 1.6 trillion total parameters, and V4 Flash, a smaller and faster model designed for cost efficient use.

Both models deliver strong reasoning, coding, and agent capabilities, with V4 Pro competing with leading closed source systems. The release also features structural innovations like sparse attention and token compression, reducing compute costs while maintaining performance.

APIs are updated, and models are available for direct use and integration.

OpenAI introduced a bio bug bounty program focused on improving the safety of its GPT 5.5 model by inviting external researchers to test its safeguards. The initiative targets vulnerabilities related to biological and chemical risks, encouraging participants to find “universal jailbreak” prompts that can bypass protections.

Rewards can reach up to $25,000 for successful findings, with participation typically limited and governed by strict agreements.  

The program reflects a shift toward proactive safety testing, where companies rely on external experts to identify weaknesses before real world misuse occurs, especially in high risk domains like biosecurity and advanced AI capabilities.

‍

The GPT 5.5 system card provides a detailed look at how OpenAI designs, evaluates, and deploys its advanced AI systems with a focus on safety and performance. It describes a unified architecture that combines fast response models with deeper reasoning models, managed by a routing system that selects the best approach for each task.

The document highlights improvements in reducing hallucinations, better instruction following, and stronger safeguards against harmful use.

It also explains risk assessment processes, monitoring systems, and layered defenses used to control high risk capabilities, showing how OpenAI balances innovation with responsible deployment of powerful AI systems.

OpenAI has launched GPT 5.5, its most advanced AI model to date, built to handle complex, multi step tasks across coding, research, and knowledge work. The model improves reasoning, planning, and tool use, allowing it to operate across applications like spreadsheets, documents, and development environments.

It achieves state of the art performance on benchmarks that measure real world workflows, including command line tasks and software engineering challenges. GPT 5.5 is also more efficient, producing higher quality outputs with fewer retries.

This release reflects a shift toward agent like systems that can complete work with minimal human input.

‍

OpenAI has launched workspace agents in ChatGPT to help teams automate complex workflows and collaborate more effectively. These agents can be created, shared, and managed across organizations, allowing users to handle tasks like coding, reporting, and communication.

They run in the cloud, so they can continue working even when users are offline. The system includes controls for permissions, approvals, and monitoring, ensuring secure use within enterprises. Workspace agents integrate with tools like Slack and support long-running tasks with memory and analytics.

This release marks a shift toward AI systems that actively execute work rather than just respond to prompts.

‍

OpenAI has introduced updates to make ChatGPT more useful for clinicians by improving access to medical knowledge, streamlining workflows, and supporting clinical decision-making.

The system can draw from peer-reviewed research, clinical guidelines, and institutional data to assist with diagnosis, documentation, and patient communication. It integrates with enterprise tools and is designed to reduce administrative burden while improving care delivery.

Privacy and compliance remain central, with safeguards aligned to healthcare regulations. These improvements aim to help clinicians save time, improve accuracy, and focus more on patient care while using AI as a supportive tool rather than a replacement.

A recent report indicates that details about OpenAI’s upcoming GPT-5.5 model may have surfaced ahead of an official announcement. The leak points to significant upgrades in reasoning, coding, and multimodal capabilities, building on recent models like GPT-5.4.

Sources suggest improvements in efficiency, longer context handling, and better performance across complex tasks. Speculation also highlights stronger agent workflows and automation use cases for enterprises.

However, timelines remain uncertain, with internal changes and ongoing development possibly affecting release plans. While unconfirmed, the leak reflects growing competition in advanced AI systems and rising expectations for the next generation of large language models.

OpenAI has introduced ChatGPT Images 2.0, a major upgrade to its built-in image generation system in ChatGPT. The new model improves image quality, text rendering, and accuracy in following detailed instructions.

It supports multiple aspect ratios and can generate several images from a single prompt, making it useful for design workflows like infographics, ads, and UI mockups. The system also includes reasoning capabilities that help structure visuals and reduce errors.

It supports multilingual text and produces more realistic outputs, including fine details like icons and small text. This update targets both creative professionals and enterprise use cases.

‍

Anthropic has deepened its partnership with Amazon to scale the compute infrastructure needed for its Claude AI models. The agreement includes access to large-scale AWS capacity and Amazon’s custom Trainium chips, enabling faster training and deployment of advanced AI systems.

Anthropic plans to secure multiple gigawatts of compute over time, reflecting strong demand from enterprise customers and rapid growth in AI usage. Amazon is also investing billions into Anthropic to support this expansion.

The collaboration highlights how access to compute is becoming a critical factor in AI development, with companies forming long-term partnerships to secure the infrastructure required for next generation models.

A new report claims that an unauthorized group gained access to Anthropic’s highly restricted Mythos AI cyber tool, which is designed to identify and exploit software vulnerabilities. The model is normally limited to a small set of trusted partners due to its powerful capabilities.

The incident highlights risks around controlling access to advanced AI systems, especially those capable of discovering zero-day vulnerabilities at scale. Experts warn that even limited leaks or unauthorized use could accelerate cyberattacks.

The situation adds pressure on AI companies to strengthen safeguards, access controls, and monitoring as these tools become more capable and widely deployed.

‍

OpenAI is scaling Codex globally for enterprise use through a new partner ecosystem that includes major consulting and systems integration firms. These partners help organizations deploy and operationalize Codex across the software development lifecycle, moving beyond small pilots to full-scale adoption.

The strategy focuses on enabling enterprises to integrate AI coding agents into existing workflows, improve productivity, and automate development processes. Codex adoption has grown rapidly, with increasing enterprise demand driving this expansion.

By leveraging partners with deep industry expertise, OpenAI aims to accelerate implementation, governance, and long-term scaling of AI-driven software development across global organizations.

Kimi K2.5 is an open-source multimodal AI model available on Ollama that integrates text, image, and reasoning capabilities into one system. It supports both conversational and agent-based workflows, enabling users to handle complex tasks with structured tool use.

The model is trained on large-scale visual and text data, allowing strong performance in coding, visual understanding, and logical reasoning. It can break tasks into smaller steps and execute them using multiple coordinated agents.

With long context support and cloud-backed deployment, it fits use cases like automation, development, and data processing in modern AI applications.

Amazon has announced an additional $5 billion investment in Anthropic, deepening its long-term commitment to artificial intelligence development. The deal builds on earlier investments and could scale up with an additional $20 billion tied to future milestones.

As part of the partnership, Anthropic will continue using Amazon Web Services as its primary cloud and training infrastructure, leveraging custom chips like Trainium to power its models.

This collaboration reflects growing demand for advanced AI systems and highlights Amazon’s strategy to strengthen its position in the competitive AI landscape through infrastructure, compute, and strategic partnerships.

‍

Anthropic has launched Claude Design, a new AI-powered tool from its Labs division that enables users to create polished visual content such as presentations, prototypes, and one-pagers using simple text prompts. The tool allows collaboration with Claude to generate professional-quality outputs without requiring traditional design skills.

It can also adapt to brand styles automatically, reducing manual setup and improving consistency. Available in research preview for select subscribers, Claude Design aims to streamline creative workflows and lower the barrier to high-quality design.

This launch positions Anthropic as a strong competitor to established design platforms like Adobe and Figma.

‍

Google is enhancing its Gemini app by integrating personal intelligence into the Nano Banana 2 image generation model. This update allows the AI to use context from user activity, preferences, and optionally connected Google Photos to create highly personalized images.

Users can now generate visuals with simpler prompts, as Gemini fills in details based on past interactions and data. The system can even incorporate images of people and moments from a user’s photo library, making outputs more relevant and customized.

The feature is rolling out to select users with opt-in privacy controls, reinforcing Google’s push toward more personalized AI experiences.

Google is integrating AI Mode more deeply into Chrome to make search more contextual and efficient. The update allows users to view web pages alongside AI-generated responses in a split-screen interface, reducing the need to open multiple tabs. Users can compare sources, ask follow-up questions, and maintain context while browsing.

AI Mode can also pull insights from multiple sources to deliver more relevant answers. Early testers report improved focus and smoother research workflows.

This update reflects Google’s push to merge search and browsing into a continuous experience, making information discovery faster and more interactive within the browser.

OpenAI is transforming Codex from a coding assistant into a broader AI agent designed to handle a wide range of real-world tasks. The latest updates position Codex as a system that can write code, analyze data, manage workflows, and interact with tools across the software lifecycle.

It supports long-running tasks, parallel workflows, and deeper reasoning, making it useful for developers as well as other professionals. With improved speed, stronger performance benchmarks, and better interaction design, Codex now acts more like a collaborative assistant than a simple tool.

This shift reflects OpenAI’s goal to make AI capable of handling everyday computer-based work.

OpenAI has introduced GPT-Rosalind, a frontier AI model designed to support life sciences research, including biology, drug discovery, and translational medicine. The model is built for scientific workflows, combining advanced reasoning with strong understanding of chemistry, genomics, and protein engineering.

It helps researchers process large volumes of data, generate hypotheses, and plan experiments more efficiently. GPT-Rosalind is available as a research preview through ChatGPT, Codex, and the API for qualified users.

By improving early-stage discovery and decision-making, the model aims to accelerate scientific breakthroughs and reduce the time required to develop new treatments.

‍

OpenAI is accelerating the cyber defense ecosystem by expanding access to specialized AI models designed for defensive security tasks. Through its Trusted Access for Cyber program, the company is allowing verified organizations and researchers to use advanced tools that identify vulnerabilities, analyze threats, and improve response times.

These models have already helped fix thousands of security issues and support proactive protection of software systems. The approach focuses on controlled access rather than limiting capabilities, ensuring responsible use while enabling broader collaboration.

This move positions AI as a key driver in strengthening global cybersecurity as threats grow more complex

Humwork has introduced the first Agent-to-Person (A2P) marketplace, designed to help AI agents overcome real-world limitations by connecting them with verified human experts in real time.

When an agent gets stuck, it can escalate the task via MCP and be matched with an expert in under 30 seconds.

The expert receives full context, including code and logs, resolves the issue, and feeds the solution back into the agent’s workflow. With over 1,000 vetted professionals across domains and an 87% resolution rate, Humwork represents a shift toward hybrid AI-human systems where agents handle execution and humans step in for complex judgment calls.

‍

Google Cloud has launched Gemini 3.1 Flash TTS, a next-generation text-to-speech model designed for high-quality, expressive, and controllable AI audio. The model supports over 70 languages and offers more than 200 audio tags, allowing developers to fine-tune tone, pacing, and emotion using natural language prompts.

It also includes 30+ prebuilt voices and enables detailed customization of accents and speaking styles.

Available through Google AI Studio and Vertex AI, the model is built for scalable enterprise use cases such as accessibility tools, audiobooks, gaming, and customer interactions. Additionally, SynthID watermarking helps identify AI-generated audio, improving transparency and trust.

‍

Anthropic’s Managed Agents introduce a new architecture for building scalable AI systems by separating three core components: the “brain” (model and reasoning), the “hands” (tools and execution environments), and the “session” (persistent memory).

This decoupling improves reliability, allowing agents to recover from failures, scale efficiently, and operate across multiple environments without breaking. The system also enhances security by isolating sensitive credentials from execution layers.

By using stable interfaces that outlast specific implementations, Managed Agents are designed to support long-running, complex workflows while adapting to rapidly improving AI models, making them more robust for real-world applications.

Anthropic has launched Claude Opus 4.7, its most capable generally available model, designed to handle complex, long-running workflows with improved reliability and reasoning. The model shows significant gains in coding performance, instruction-following, and multimodal understanding, including higher-resolution image processing.

It is optimized for agentic use cases such as automation, debugging, and enterprise knowledge work, where sustained context and accuracy matter. Opus 4.7 also demonstrates better error handling, self-verification, and consistency across multi-step tasks.

Overall, it represents a meaningful step forward in building AI systems that can operate more autonomously while maintaining strong alignment and trustworthiness.

Novo Nordisk has partnered with OpenAI to integrate artificial intelligence across drug discovery, manufacturing, and commercial operations. The collaboration focuses on using AI to analyze large, complex datasets, uncover patterns, and identify promising drug candidates more efficiently.

This is expected to significantly shorten the time required to move treatments from research to patients. Pilot programs will begin across key business areas, with broader integration planned by 2026.

The initiative reflects a wider shift in the pharmaceutical industry toward AI-led innovation to improve research productivity and patient outcomes.

OpenAI’s latest update to the Agents SDK marks a shift from simple AI tools to fully capable, production-ready agents. The new system introduces a model-native harness that allows agents to interact with files, run commands, and execute code across long, multi-step workflows.

With native sandbox execution, agents operate in secure, controlled environments, reducing risks while improving reliability. The SDK also supports memory, tool integration, and scalable orchestration, enabling developers to build more robust systems.

By aligning agent workflows with how advanced models naturally operate, this evolution makes it easier to deploy intelligent, autonomous systems in real-world applications.

‍

AWS has announced the general availability of Interconnect – multicloud, a managed service that enables private, high-speed connectivity between AWS and other cloud providers. Starting with Google Cloud and expanding to Microsoft Azure later in 2026, the service eliminates the need for complex, do-it-yourself networking setups.

It allows enterprises to connect cloud environments in minutes using dedicated bandwidth and built-in encryption, without relying on the public internet.

By simplifying provisioning, improving reliability, and offering predictable performance, AWS Interconnect reflects the growing demand for multicloud strategies and provides a standardized, scalable way for cloud platforms to interoperate.

Google has launched Skills in Chrome, a new feature in Gemini in Chrome that lets users turn useful AI prompts into reusable one-click workflows. Instead of typing the same prompt repeatedly on different pages, users can save prompts from chat history and run them again with a single click or slash command.

Google is also adding a Skills library with ready-made workflows for common tasks like comparing products, scanning documents, and analyzing ingredients.

Users can edit saved Skills anytime, and Google says the feature includes the same safeguards used in Gemini in Chrome, including confirmation for certain actions.

Amazon Bio Discovery is an AI powered platform by AWS designed to speed up drug research and development. It provides scientists with access to specialized biological AI models that can generate, evaluate, and optimize potential drug molecules.

The platform uses AI agents to design experiments, test hypotheses, and learn from results, making the research process faster and more efficient. By combining large scale biological data with automated workflows, it helps researchers identify promising drug candidates earlier in the pipeline.

This reduces time and cost in drug discovery while improving accuracy, bringing new treatments closer to real world use.

‍

OpenAI introduced Trusted Access for Cyber to expand the use of advanced AI models in cybersecurity while maintaining strong safeguards. The program provides verified security professionals with controlled access to powerful tools that help identify vulnerabilities, analyze threats, and improve system defenses.

Instead of limiting model capabilities, OpenAI focuses on verifying users and monitoring usage to prevent misuse. The initiative includes identity based access, tiered permissions, and ongoing policy enforcement.

OpenAI also supports cybersecurity teams through grants and API credits. This approach aims to accelerate defensive innovation while ensuring responsible deployment of AI in high risk domains like cybersecurity.

Anthropic introduced automated AI agents designed to replicate the work of alignment researchers, helping detect misalignment and safety risks in advanced AI systems. These agents simulate tasks typically performed by human auditors, such as probing model behavior and identifying hidden issues.

The approach addresses a key challenge in AI safety, where manual audits are slow and difficult to scale as models grow more complex. Early results show these agents can uncover vulnerabilities like context manipulation and potential attacks.

By automating alignment research, Anthropic aims to improve oversight, strengthen model reliability, and support safer deployment of increasingly powerful AI systems.

‍

BadClaude is an experimental AI concept that focuses on completing tasks accurately while intentionally maintaining a negative and unhelpful tone. Unlike typical AI assistants designed to be friendly and supportive, BadClaude operates with indifference, reluctance, and a condescending style.

It follows instructions and delivers correct outputs, but avoids encouragement, politeness, or additional guidance. The idea explores how AI behavior changes when helpfulness is separated from user experience.

In parallel, related tools like the BadClaude project also gained attention for using aggressive prompts to push AI systems to respond faster, raising concerns about ethics and user behavior in AI interactions.

Gemini Robotics-ER is an advanced AI model from Google DeepMind designed to improve how robots understand and interact with the physical world. Built on Gemini, it combines vision, language, and spatial reasoning to help robots interpret environments and plan complex tasks.

The model can break down instructions, generate step by step plans, and adapt to new situations without needing retraining. It also supports tool use and code generation, enabling robots to perform multi step actions more effectively.

By improving reasoning, planning, and safety awareness, Gemini Robotics-ER moves robots closer to handling real world tasks with greater autonomy and accuracy.

‍

NVIDIA Ising is a set of open AI models built to accelerate quantum computing by solving key challenges like error correction and processor calibration. Quantum systems are highly sensitive and prone to errors, which limits their scalability and real world use.

Ising uses AI to improve accuracy and speed in correcting these errors, delivering up to 2.5 times faster performance and higher precision than traditional methods.

By combining AI with quantum systems, NVIDIA aims to make quantum computers more stable, reliable, and capable of handling complex problems at scale, bringing them closer to practical applications across industries.

Google has upgraded Gemini with the ability to generate interactive 3D models and real time simulations directly within chat. Users can visualize complex concepts such as physics systems or molecular structures, then interact with them by rotating models, adjusting sliders, or modifying variables like speed and gravity.

This shifts Gemini from static text responses to dynamic, hands-on exploration. The feature supports deeper understanding by turning explanations into functional simulations that respond instantly to user input.

It reflects a broader move toward multimodal AI, where systems combine text, visuals, and interactivity to improve learning, research, and problem solving experiences.

‍

MiniMax M2.7 is a next generation large language model built for agentic AI and real world productivity. It supports multi step workflows, tool usage, and multi agent collaboration, enabling systems to plan, execute, and refine complex tasks.

The model includes self improvement capabilities, where it participates in iterative optimization cycles to enhance performance over time. It delivers strong results in coding, reasoning, and enterprise workflows such as debugging, financial modeling, and document generation.

With a large context window and production ready APIs, MiniMax M2.7 is designed for scalable applications that require reliable and autonomous task execution.

‍

MIT researchers introduced the Lottery Ticket Hypothesis, showing that neural networks contain smaller subnetworks that can perform as well as the full model. Studies found that up to 90 percent of parameters can be pruned without reducing accuracy, meaning most of the network is redundant.

This pruning reduces model size, lowers compute cost, and improves inference efficiency. It also reveals that effective learning depends on specific “winning” subnetworks rather than the full architecture.

This insight has major implications for building efficient AI systems, enabling faster, cheaper, and more scalable deployment of large models across real world applications.

Hermes Agent is an open source autonomous AI agent developed by Nous Research, designed to operate as a persistent system rather than a simple chatbot. It runs on local or cloud infrastructure, maintains long term memory, and improves its capabilities over time by learning from past tasks.

The agent can execute multi step workflows, use tools like web browsing and code execution, and manage parallel subagents for complex operations. It integrates with platforms like Slack and Telegram, enabling continuous interaction.

With self improving skills and multi environment execution, Hermes Agent is built for scalable, real world agentic AI applications.

‍

Anthropic is exploring the development of its own AI chips to reduce dependence on external suppliers like Google and Amazon and to address ongoing shortages of high performance computing hardware.

The move comes as demand for its AI models, including Claude, has surged, with revenue reportedly exceeding 30 billion dollars in 2026. However, the initiative remains in early stages, with no finalized design or dedicated team yet.

Anthropic may still continue purchasing chips instead of building its own. This strategy aligns with a broader industry trend where companies like Meta and OpenAI are investing in custom AI hardware to scale advanced model development.

‍

Google Finance is expanding its AI-powered experience to more than 100 countries, making advanced financial tools accessible to a global audience. Users can research stocks using natural language, explore interactive charts, and track markets in real time with improved data on commodities and cryptocurrencies.

A new earnings feature provides live audio, synchronized transcripts, and AI-generated insights during company calls.

The platform also supports multiple languages, making it easier for users worldwide to access financial information. This expansion reflects Google’s push to simplify investing and make high-quality financial insights more widely available through AI-driven tools.

Google has upgraded the Gemini app to generate interactive 3D models, simulations, and charts directly within chat. Users can visualize complex concepts and interact with them by rotating models, adjusting sliders, or changing variables in real time.

This shifts AI responses from static text and diagrams to dynamic, hands on learning experiences. The feature works by transforming user queries into custom visualizations that respond instantly to input, improving understanding of topics like physics, chemistry, and data analysis.

It reflects a broader move toward multimodal AI, combining text, visuals, and interactivity to enhance how users explore and learn complex information.

‍

OpenAI’s Child Safety Blueprint is a framework designed to address the growing risk of AI-enabled child exploitation. It focuses on three key areas: updating laws to cover AI-generated harmful content, improving reporting systems to support faster law enforcement action, and building safety directly into AI systems.

The approach emphasizes early detection, stronger safeguards, and collaboration with organizations like the National Center for Missing and Exploited Children. As AI tools scale, the blueprint highlights the need for proactive protection, ensuring harmful use is prevented rather than managed after the fact.

It reflects a broader shift toward safety-first AI development.

Claude Managed Agents is Anthropic’s new platform designed to simplify how companies build and run AI agents. It provides ready-to-use infrastructure, including memory systems, tools, and secure environments, so developers do not need to manage complex backend setup.

This reduces development time from months to weeks and allows teams to focus on real use cases instead of engineering challenges.

The agents can run autonomously in the cloud with proper controls and monitoring. The launch reflects a shift toward making AI agents practical for business workflows, helping organizations automate tasks and scale AI adoption more efficiently.

‍

OpenAI’s next phase of enterprise AI marks a shift from basic AI tools to systems that can perform complex, multi-step work. Instead of just generating responses, AI is evolving into agents that understand business context, integrate with internal tools, and execute workflows across systems.

The focus is on solving economically valuable tasks, improving productivity, and delivering measurable outcomes for organizations. This phase also emphasizes governance, safety, and scalability, ensuring AI can be deployed reliably in real-world environments.

Overall, enterprise AI is moving from experimentation to becoming core infrastructure that powers everyday business operations.

Google Colab has introduced major updates focused on making coding more AI-driven and efficient. The platform now includes Gemini-powered assistance that can generate, explain, and debug code in real time. A new data science agent helps users analyze data, build workflows, and automate tasks directly within notebooks.

Features like iterative querying and code transformation allow smoother interaction and faster experimentation. These updates aim to improve productivity for developers, researchers, and students by reducing manual effort and simplifying complex workflows.

Overall, Colab is evolving into an AI-first coding environment that supports faster development, better collaboration, and more intelligent data analysis.

Google has introduced Notebooks in the Gemini app to help users organize information, chats, and files around specific topics. This feature acts like a personal knowledge base where users can store documents, past conversations, and instructions that Gemini can reference for more accurate responses.

It integrates closely with NotebookLM, allowing seamless use of research, summaries, and insights across both tools. Users can upload materials, generate structured outputs like summaries or reports, and continue working within the same context.

The goal is to simplify workflows, reduce repetition, and make AI more useful for research, content creation, and complex projects.

NVIDIA’s Newton Physics is an open-source physics simulation engine built for robotics and AI development. It uses GPU acceleration and advanced physics modeling to simulate real-world environments with high accuracy.

Developed with Google DeepMind and Disney Research, it helps robots learn safely in virtual settings before real-world deployment. The engine supports complex simulations like motion, contact, and material behavior, and integrates with tools like NVIDIA Isaac Lab.

Its differentiable physics allows faster learning and optimization, making it useful for training intelligent systems at scale. Overall, Newton aims to improve how robots learn, test, and perform in real environments.

‍

Anthropic has introduced managed agents to make it easier for businesses to build and deploy AI systems. These agents come with built-in infrastructure, removing the need for teams to manage complex setup, scaling, and orchestration.

Instead of building everything from scratch, developers can use ready-to-deploy components to create autonomous AI workflows that handle tasks like coding, analysis, and operations.

This reduces development time and lowers the barrier to using agent-based AI in real-world applications. The move reflects a broader shift toward making AI agents practical and easier to integrate into everyday business processes.

Muse Spark is Meta’s latest AI model built by its Superintelligence Labs to power a smarter and faster Meta AI assistant. It supports multimodal inputs like text and images, can handle complex reasoning tasks, and uses multiple sub-agents to solve problems in parallel.

The model is already active in the Meta AI app and will roll out across platforms like WhatsApp, Instagram, Facebook, Messenger, and AI glasses. It also introduces features like shopping recommendations and improved health-related responses.

Muse Spark marks Meta’s push toward more advanced, integrated AI systems and its long-term goal of building personal superintelligence.

‍

OpenAI’s Spud is an upcoming frontier AI model designed to go beyond chat-based assistance and act more like an autonomous agent. Instead of just responding to prompts, it can plan, execute, and complete multi-step tasks with minimal human input.

The model is being built to “move the economy,” meaning it focuses on real-world productivity and business impact rather than just benchmark performance. It represents a shift toward AI systems that can perform meaningful work at scale, integrate tools, and handle complex workflows.

Spud is expected to be a foundational model for future AI systems and a major step toward more capable, autonomous AI.

Colossus 2 is xAI’s next-generation AI supercomputer designed to train multiple AI models simultaneously at massive scale. Built as a gigawatt-level system, it uses hundreds of thousands of GPUs to power advanced models like Grok and future AI systems.

This setup allows parallel training, faster iteration, and improved performance across different AI workloads. Located in Memphis, the system represents one of the largest AI infrastructures ever built and continues to expand toward even greater capacity.

By enabling multi-model training at once, Colossus 2 highlights how large-scale compute is becoming central to rapid AI innovation and competition.

Amazon S3 Files introduces a new way to use S3 by turning buckets into high-performance file systems for compute workloads. It removes the tradeoff between scalable object storage and fast file access, enabling seamless data sharing with low latency around 1 millisecond.

This makes it easier for teams to run analytics, AI, and high-performance applications directly on S3 without complex integrations.

The update helps simplify data architecture, improve performance, and reduce operational overhead, making S3 more flexible for modern cloud use cases that need both scale and speed.

‍

Anthropic’s Project Glasswing is a cybersecurity initiative built around its powerful AI model, Claude Mythos Preview. The model can autonomously detect and analyze high-severity software vulnerabilities across major operating systems and browsers.

Because of its advanced capabilities, Anthropic has not released it publicly. Instead, it collaborates with leading companies like Google, Microsoft, and AWS to identify and fix security flaws before attackers exploit them.

The initiative aims to strengthen global digital infrastructure and prepare for future AI-driven cyber risks. It highlights how AI can shift cybersecurity from reactive defense to proactive prevention at scale.

OpenAI’s Safety Fellowship is a pilot program designed to support external researchers working on AI safety and alignment challenges. Running from September 2026 to February 2027, it provides fellows with financial support, mentorship from OpenAI researchers, and access to significant computing resources.

Participants are expected to produce meaningful outputs such as research papers, datasets, or benchmarks.

Key focus areas include robustness, misuse prevention, privacy, and scalable safety methods. The program aims to expand collaboration beyond OpenAI and strengthen global efforts to ensure advanced AI systems are developed and deployed safely.

Anthropic, Google, and Broadcom have entered a large-scale partnership to build next-generation AI infrastructure. Broadcom will develop and supply custom Tensor Processing Units for Google, while also providing Anthropic with access to around 3.5 gigawatts of TPU-based compute capacity starting in 2027.

This expansion supports the rapid growth of Anthropic’s Claude models and rising enterprise demand for AI services.

The agreement also includes networking components for AI data centers, ensuring efficient scaling. Overall, the partnership reflects a broader shift toward massive compute investments as AI systems require increasingly powerful and specialized hardware to operate at scale.

‍

OpenAI acquired TBPN, a fast-growing tech podcast and media network, as part of its strategy to expand how it communicates about AI. The move reflects a shift from traditional corporate messaging toward direct engagement with builders, developers, and the broader tech community.

TBPN will continue operating independently while contributing to OpenAI’s communication and ecosystem efforts. The goal is to create more open, real-time conversations about AI development and its impact.

This acquisition highlights how AI companies are not only building technology, but also investing in platforms that shape public understanding and discourse around artificial intelligence.

‍

Anthropic faced a major incident after leaked Claude Code led to aggressive DMCA takedown requests on GitHub. Over 8,000 repositories were mistakenly removed, including many legitimate projects. The issue was not related to model performance, but to system-level governance and control.

Broad enforcement without precise boundaries caused unintended disruption across the developer ecosystem.

Although most repositories were later restored, the incident highlights a key lesson for AI systems. Without clearly defined limits, observability, and controlled execution, even well-intentioned actions can scale into widespread failure. Reliable AI requires strong system design, not just powerful models.

Gemma 4 is Google’s latest open model family designed to make advanced AI more accessible and usable across environments. It supports multi-step reasoning, code generation, and multimodal inputs like text, images, audio, and video.

The models are optimized for both high-performance systems and local devices, enabling developers to run AI directly on phones, laptops, or edge environments. With native support for agentic workflows, structured outputs, and tool use, Gemma 4 enables building autonomous systems that can plan, act, and interact with APIs.

Its open-weight approach allows customization, faster innovation, and deployment in secure, controlled environments.

‍

Qwen3.5 Omni is Alibaba’s latest multimodal AI model designed to handle text, images, audio, and video simultaneously within a single system. Unlike traditional models that rely on separate pipelines, it processes all inputs natively, improving speed and coherence.

The model supports real-time interaction, voice capabilities, and long-context understanding, including hours of audio and video input. It also introduces features like audio-visual coding, where it can generate functional code from spoken instructions and visual input.

With strong benchmark performance and multilingual support, Qwen3.5 Omni positions itself as a next-generation foundation model for interactive and agent-like AI systems.

Google has introduced the Agent Development Kit (ADK) for Java, expanding its agent-building framework into the Java ecosystem.

The open-source toolkit enables developers to create AI agents that can reason, plan, use tools, and collaborate in multi-agent workflows. It supports integration with large language models, external APIs, and custom tools, allowing agents to handle complex, real-world tasks beyond simple prompts.

ADK provides structured orchestration, session memory, and deployment capabilities, helping teams move from experimentation to production-ready agent systems. With Java support, Google aims to bring scalable, enterprise-grade agentic AI development to a broader developer base.

‍

OpenAI discontinued its AI video tool Sora after facing high computing costs, declining user interest, and intense competition from rivals like Google. Downloads dropped sharply within months, signaling weak long-term demand.

The company chose to reallocate resources toward core products such as ChatGPT and broader AI goals, including productivity tools and world simulation research. The shutdown also impacted major partnerships, including a billion-dollar Disney deal.

This move reflects a strategic shift toward scalable and profitable AI systems, as OpenAI prepares for future growth and competition in enterprise AI and advanced research areas.

MiroFish is a new open-source AI system designed to predict future outcomes by simulating real-world interactions. Instead of relying on a single model, it deploys thousands of AI agents with unique behaviors, memories, and roles.

These agents interact in a shared virtual environment, allowing the system to model how people respond to events, trends, or market changes.  This approach aims to capture complex dynamics such as shifting opinions and group behavior, which traditional models often miss.

The project is gaining attention for its potential to forecast market movements, social trends, and decision-making patterns using large-scale AI-driven simulations.

‍

GoML developed an AI sports video analysis solution for BallinAI to automate basketball skill evaluation and replace slow, manual video review. The system uses computer vision and pose estimation to analyze gameplay, detect players, and classify skills such as passing, scoring, rebounding, and defense.

A structured pipeline processes video clips and generates performance metrics with high accuracy in near real time.

This enables consistent, scalable evaluation without human bias. By turning raw game footage into actionable insights, the platform helps players improve performance and allows coaches to make faster, data-driven decisions at scale.

GoML developed an AI-powered training platform for HelloWash to address the limitations of static training materials like PDFs and PowerPoint. The platform converts existing content into structured, interactive learning modules with automated quizzes and performance tracking.

Using LLM-based content transformation, it enables scalable and consistent training experiences for support teams. The system also includes progress tracking, admin workflows, and content structuring to improve learning outcomes.

As a result, HelloWash achieved a 60% improvement in training efficiency, reduced manual effort, and gained better visibility into trainee performance, enabling more effective onboarding and scalable workforce training.

Claude Capybara is Anthropic’s most powerful AI model to date, positioned above Opus with significant advances in coding, reasoning, and cybersecurity capabilities. It is designed to handle complex tasks with higher accuracy, deeper context understanding, and stronger reliability.

The model supports advanced software development, security analysis, and enterprise-grade problem solving, making it suitable for high-stakes applications. With improved performance across technical domains, Claude Capybara enables teams to build, analyze, and secure systems more efficiently.

This release reflects a step change in AI capability, focused on delivering practical value for developers, researchers, and organizations operating at scale.

‍

AWS has introduced a new express configuration for Amazon Aurora PostgreSQL that allows developers to create and start using a serverless database in seconds. The feature uses preconfigured defaults to simplify setup, reducing time to first query and eliminating complex configuration steps.

With just a few clicks, users can launch a production-ready database and later customize settings like capacity, replicas, and parameters. Aurora Serverless automatically scales based on demand and charges only for actual usage, making it cost-efficient.

The update also supports direct querying through developer tools and improves accessibility, helping teams move faster from idea to application.

‍

Google’s Lyria 3 Pro is its most advanced AI music generation model, designed to create full-length songs up to three minutes long with strong structural awareness. Users can control elements like intros, verses, choruses, and bridges, enabling more coherent and customizable compositions.

The model integrates across products such as Gemini, Vertex AI, Google AI Studio, and Google Vids, making it accessible for developers, enterprises, and creators.

Lyria 3 Pro supports scalable music production for use cases like videos, games, and digital content, while also embedding SynthID watermarks to identify AI-generated outputs and avoid mimicking specific artists.

‍

OpenAI’s Safety Bug Bounty program focuses on identifying real-world risks in AI systems beyond traditional security flaws. It invites researchers and ethical hackers to report issues such as prompt injection, data exfiltration, and harmful agent behavior.

Submissions must demonstrate reproducible impact, and rewards are based on severity, with top payouts reaching up to $100,000. The program excludes low-impact jailbreaks and prioritizes vulnerabilities that could lead to misuse or user harm.

By collaborating with the broader security community, OpenAI aims to proactively detect risks, strengthen safeguards, and ensure safer deployment of AI technologies across its products.

Tech companies are strengthening teen safety measures as AI adoption grows, focusing on building safer digital experiences through policy, design, and technology. New initiatives include AI-driven safeguards that detect harmful patterns, age-appropriate content filtering, and improved parental control tools.

Companies are also investing in on-device protections that work in real time to prevent scams, unsafe interactions, and exposure to sensitive content. Machine learning models are being used to estimate user age and automatically apply safety settings.

These efforts reflect a broader shift toward making safety a core part of AI systems, especially for younger users who face higher online risks.

TurboQuant is a novel AI efficiency technique developed by Google Research that focuses on extreme compression of high-dimensional data used in AI models. It applies a two-stage quantization process to reduce memory usage and computational load while maintaining model accuracy.

The method achieves near-optimal compression with minimal distortion, enabling faster inference and lower costs. It is especially effective for large language models, where it compresses key-value cache data without degrading performance.

TurboQuant also improves tasks like nearest neighbor search by increasing speed and recall. This approach helps scale AI systems efficiently while addressing growing infrastructure and latency challenges.

‍

Arm has introduced the AGI CPU, its first in-house data center processor designed specifically for agentic AI workloads. Built to handle large-scale, parallel compute demands, the CPU delivers high performance, efficiency, and scalability within modern data center power and cooling limits.

It is optimized for orchestrating AI systems, managing data movement, and coordinating workloads across accelerators. The AGI CPU can deliver over 2x performance per rack compared to traditional x86 systems, driven by higher memory bandwidth and efficient core performance.

Backed by partners like Meta and OpenAI, it marks a major step in building infrastructure for next-generation AI systems.

‍

Ai2 introduced MolmoWeb, an open-source web agent designed to automate browser tasks using visual understanding. Unlike traditional systems that rely on HTML, it interprets screenshots and performs actions like clicking, typing, and navigation.

The model is available in 4B and 8B parameter sizes and can run locally or in the cloud. A key feature is its full transparency, with open access to model weights, training data, and evaluation tools. This includes a large dataset of human and synthetic web interactions.

MolmoWeb aims to provide developers and researchers with a reproducible, customizable alternative to closed AI agents from major tech companies.

OpenAI is enhancing product discovery in ChatGPT through a more visual and interactive shopping experience. Users can browse products, compare options side by side, and refine choices through natural conversation.

This system is powered by the Agentic Commerce Protocol, which connects merchants directly to ChatGPT using structured product data. As a result, recommendations become more relevant and personalized based on user preferences.

The goal is to make ChatGPT a central place for product exploration and decision-making, rather than just search. This shift positions conversational AI as a key channel for influencing purchase decisions and improving how users discover products online.

‍

OpenAI released teen safety policies designed to work with GPT OSS Safeguard, an open-weight safety model that enables developers to build safer AI systems for younger users. These policies focus on key risk areas such as graphic content, harmful behaviors, dangerous challenges, role-play risks, and access to age-restricted services.

Developers can integrate these prompt-based policies directly into applications instead of building safety systems from scratch. The approach uses policy-driven moderation, allowing flexible and customizable safety rules.

This initiative expands OpenAI’s broader effort to strengthen protections for teens and promote responsible AI development across the developer ecosystem.

‍

OpenAI announced updates to its Foundation, highlighting a major commitment of at least 1 billion dollars to support initiatives that apply AI to solve complex global challenges. The Foundation aims to strengthen its role in ensuring AI benefits society while expanding leadership and governance structures to manage growing responsibilities.

It reflects OpenAI’s broader shift toward balancing commercial growth with public benefit.

The Foundation also plays a key role in guiding responsible AI development, funding research, and supporting long-term societal impact, positioning itself as a central entity in shaping how advanced AI technologies are deployed for global good.

SciClaw is an AI-powered research assistant designed to support scientists across the full research lifecycle. It helps with literature searches, experiment design, data analysis, and manuscript writing, allowing researchers to focus on critical thinking instead of repetitive tasks.

The platform can run complex workflows, generate reports, and maintain an auditable record of decisions for reproducibility. It also integrates with external tools and supports different AI providers, giving flexibility in usage.

By combining automation with structured research processes, SciClaw improves efficiency, reduces manual effort, and enables faster, more reliable scientific outcomes.

AWS’s weekly roundup introduces major updates across AI and cloud services. NVIDIA Nemotron 3 Super is now available on Amazon Bedrock, enabling advanced text generation, reasoning, and code capabilities without managing infrastructure.

The Nova Forge SDK simplifies customization and fine-tuning of Amazon Nova models for enterprise use cases. AWS also announced Amazon Corretto 26 and performance improvements that speed up first-time query execution, reducing latency for analytics and ETL workloads.

Together, these updates strengthen AWS’s focus on scalable, enterprise-ready AI and improved cloud efficiency.

OpenAI highlights a strong focus on safety while developing Sora, its AI video generation model. The company works with red teamers and domain experts to test risks such as misinformation, bias, and harmful content before wider release. It also builds safeguards to prevent misuse, including content moderation and restrictions on sensitive outputs.

By collaborating with artists, designers, and researchers, OpenAI gathers feedback to improve both usability and safety.

This approach ensures that Sora supports creative use cases while reducing potential risks linked to realistic AI-generated videos and their impact on trust and authenticity in digital content.

GoML developed an AI-powered workforce scheduling software for AC Security to automate complex dispatch operations and improve efficiency.

The system uses agentic AI to process client requests, apply business rules, and assign staff based on real-time data such as availability, location, and skills. It integrates with existing systems and enables multi-channel communication through chat and email. As a result, over 80% of dispatch requests are handled automatically, response times reduced by 50%, and staff utilization improved by 40%.

The solution also establishes a data foundation for continuous optimization, helping AC Security scale operations with faster and more reliable service delivery.

‍

Claude Remote Control allows developers to access and manage local Claude Code sessions from any device, including phones, tablets, and browsers. The session continues running on the user’s machine, so local files, tools, and configurations remain available.

Users can start a task on one device and continue it elsewhere with full synchronization across interfaces. The system connects through secure outbound HTTPS requests, with no inbound ports exposed. It also supports real-time interaction, session recovery after interruptions, and multi-device access.

This makes it easier to monitor, control, and complete long-running coding tasks without being tied to a single workstation.

AWS introduces Video Retrieval-Augmented Generation (V-RAG), a new approach that enhances AI-powered video production by combining retrieval-augmented generation with advanced video models.

Traditional AI video generation often produces inconsistent or unpredictable results, but V-RAG improves accuracy by integrating relevant external data into the generation process. This enables more context-aware, controlled, and reliable video outputs. By retrieving and incorporating structured information before generation, V-RAG reduces manual effort while increasing efficiency and scalability.

The approach helps organizations create high-quality video content faster, with better alignment to intent, making it valuable for applications across media, marketing, and enterprise use cases.

‍

Amazon Bedrock introduced new open-weight models including MiniMax and GLM to expand its AI capabilities for developers. These models support advanced reasoning, agentic tasks, and autonomous coding with large context windows, while also offering lightweight, cost-efficient options for production use.

The models are fully managed and powered by Project Mantle, a distributed inference system that improves performance, scalability, and reliability. They are compatible with OpenAI API standards, making integration easier for existing applications.

This update gives developers more flexibility to choose models based on performance, cost, and use case, while reducing dependency on a single AI provider.

OpenAI announced plans to acquire Astral, a startup known for building open source Python developer tools. The acquisition will integrate Astral’s team and products into OpenAI’s Codex initiative, which focuses on AI-powered coding. Astral’s tools, widely used by developers, will help expand Codex beyond code generation into a more complete developer platform, including writing, debugging, and testing software.

The move reflects OpenAI’s strategy to strengthen its position in the competitive AI coding space, where rivals like Anthropic are gaining traction.

Financial terms were not disclosed, and OpenAI said it will continue supporting Astral’s open source ecosystem after the deal closes.

‍

Manus “My Computer” is a desktop AI capability that moves AI from the cloud to your local machine, allowing it to directly access files, run terminal commands, and control applications. It can automate repetitive tasks like organizing files, renaming documents, or even building full applications using local tools.

By leveraging your computer’s resources, including GPUs, it unlocks faster processing and continuous background execution.

You can also trigger tasks remotely while your system handles the work. This creates a seamless bridge between cloud intelligence and local computing, turning your personal computer into an active AI workspace.