Anthropic claims its newly released Claude Sonnet 4.5 is the “best coding model in the world”
Anthropic has announced the release of Claude Sonnet 4.5, which it claims is the “best coding model in the world” and the “strongest model for building complex agents.”
It achieves a 77.2% on the SWE-bench for software engineering, compared to 74.5% for Claude Opus 4.1 and 72.7% for Claude Sonnet 4. For external comparison, GPT-5 Codex scored at 74.5%, GPT-5 scored 72.8%, and Gemini 2.5 Pro scored 67.2%.
Additionally, it leads in the OSWorld benchmark, which tests AI models on real-world computer tasks. It scored 61.4% on that benchmark, beating out Claude Sonnet 4, which scored 42.2%.
“Sonnet 4.5 can produce near-instant responses or extended, step-by-step thinking that is made visible to the user,” Anthropic says.
Google adds Data Commons MCP Server, new versions of Gemini 2.5 Flash and Flash-Lite
The Data Commons MCP Server allows AI developers to easily access all of Data Commons’ publicly available datasets. It can be accessed via the Gemini CLI or in Google Colab, and Google has a sample agent in Colab as well to make it easier to get started.
The newest version of Gemini 2.5 Flash-Lite features better instruction following, more concise answers to reduce token costs, and stronger multimodal and translation capabilities. The updated Gemini 2.5 Flash offers better agentic tool use and is more efficient, leading to reductions in cost.
OpenAI adds shared projects to ChatGPT Business subscribers
Shared projects allow multiple people to add files and instructions to a project, so that ChatGPT can provide more tailored responses for everyone involved. “Members can chat with the project’s context to stay on the same page as new information gets added and create work that stays consistent in tone and style,” OpenAI explained.
The company also added new connectors for Gmail, Google Calendar, Microsoft Outlook, Microsoft Teams, SharePoint, GitHub, Dropbox, and Box. This allows ChatGPT to offer more relevant answers based on information in those tools.
Finally, ChatGPT now has ISO 27001, 27017, 27018, and 27701 certifications; an expanded SOC 2 report; role-based access controls; and enhanced SSO.
Microsoft unveils reimagined Marketplace for cloud solutions, AI apps, and more
Microsoft has restructured its Marketplace to serve as a central place for organizations to find cloud solutions, AI apps, and agents.
This new reimagining brings together Azure Marketplace and Microsoft AppSource to simplify cloud and AI management, Microsoft explained.
It includes tens of thousands of cloud and industry solutions that can help with everything from data and analytics to productivity to security. It also features more than 3,000 AI apps and agents.
CData launches Connect AI to provide agents access to enterprise data sources
CData has announced the launch of a new managed Model Context Protocol (MCP) platform bringing together AI assistants, agent orchestration, workflow automation, and embedded AI applications—combined with access to over 300 enterprise data sources.
According to the company, Connect AI preserves data semantics and relationships in enterprise data to give AI agents better context while still providing governance over that data access.
CData’s Connect AI inherits the existing security and authentication protocols set up in the source system. Data access gets logged under the identity of the authenticated user or agent, and additional controls can be layered on top and managed in Connect AI.
Snowflake and other data companies join forces to develop vendor-neutral standard for semantic metadata
A number of data companies—including Snowflake, Salesforce, BlackRock, dbt Labs, and RelationalAI—have announced the formation of a new open source initiative to create a vendor- neutral standard for defining and sharing semantic metadata.
The Open Semantic Interchange has three main goals: enhance interoperability across tools and platforms, accelerate adoption of AI and BI applications, and streamlining operations.
According to the group, organizations rely on a patchwork of AI, BI, and analytics tools, and this initiative will develop a shared semantic standard that allows these tools to “speak the same language.”
By standardizing how semantics are defined and shared, the Open Semantic Interchange hopes to ensure that data is governed, consistent, and context-rich, helping with adoption of AI.
AWS launches IDE extension for building browser automation agents
AWS has announced the launch of its open source Nova Act extension, which allows developers to build browser automation agents in their IDE, reducing the need to switch between dev and test environments.
With the new extension, developers can use natural language to describe their workflow and then the Nova Act extension will generate an agent script. That script can then be modified in a notebook-style builder, where developers can integrate APIs, data sources, and authentication, and can validate it with local testing tools.
“This extension transforms my agent development workflow by positioning Nova Act extension as a full-stack agent builder tool—a complete agent IDE for the entire development lifecycle. I can prototype with natural language, customize with modular scripting, and validate with local testing—all without leaving my IDE—ensuring production-grade scripts,” Donnie Prakoso, principal developer advocate at AWS, wrote in a blog post.
Sentry’s AI code review is now in beta
The solution uses AI to identify and fix issues in code. It will automatically flag high-impact issues in pull requests so that developers can understand where and why a bug might occur. It can also detect typos, formatting errors, and logical mistakes in pull requests. Finally, it can generate unit tests for the code in a pull request.
“The only thing easier than debugging errors with Sentry is having fewer errors to debug in the first place,” said Rohan Bhaumik, senior product manager at Sentry. “By combining predictive error detection with automated testing, AI code review dramatically reduces wasted time in code reviews, strengthens test coverage, and lets teams merge with confidence.”
OpenAI updates Codex
The company released GPT-5-Codex, a variant of GPT-5 that is optimized for Codex, OpenAI’s AI coding agent. It was trained on real-world engineering tasks like building projects from scratch, adding features and tests, debugging, large-scale refactoring, and code reviews.
“With these updates, Codex moves closer to what we’ve been building toward all along—a teammate that understands your context, works alongside you, and reliably takes on work for your team,” OpenAI wrote in a post.
Other recent updates to Codex have included the Codex CLI; the Codex IDE extension in VS Code, Cursor, and other VS Code forks; and more advanced code review capabilities.
Xcode 26 gets Claude integration
Xcode is Apple’s IDE for building apps across Apple platforms, and Claude users will now be able to connect up their Anthropic account to their Xcode environment to get access to Claude Sonnet 4 capabilities.
In Xcode, Claude can help generate documentation, provide explanations of specific sections of code, create SwiftUI previews and playgrounds, and make inline code changes in the editor.
According to Anthropic, Claude subscription usages are shared across platforms, and this integration is available for any Claude subscription that includes access to Claude Code.
GitHub launches MCP Registry to provide central location for trusted servers
GitHub has launched an MCP Registry to provide developers with a curated directory of MCP servers.
“If you’ve tried connecting AI agents to your development tools, you know the pain: MCP servers scattered across numerous registries, random repos, buried in community threads — making discovery slow and full of friction without a central place to go. Meanwhile, MCP server creators are worn out from publishing to multiple places and answering the same setup questions again and again,” GitHub wrote in a blog post.
Each server in the Registry is connected to its own GitHub repository, and they can be sorted by GitHub stars and community activity.
According to GitHub, this backing builds trust in specific MCP servers, leading to a healthier overall AI ecosystem.
Google further integrates AI into Chrome
Chrome is getting a new AI browsing assistant called Gemini in Chrome that can do things like answer questions about an article or find references in a YouTube video. It is now rolling out to U.S. Mac and Windows users who have their default language set to English, and will expand to Android and iOS in the future.
Google Search’s AI Mode will also be integrated into the Chrome address bar. For example, when a user is shopping for a mattress, it might suggest follow-up searches, such as “what’s the warranty policy?”
Finally, Google will continue using AI to keep users safe, such as filling in login credentials using Chrome’s autofill, blocking new types of scams, and helping users fix security issues like compromised passwords and spam notifications. Google says that its initial use of AI-powered warnings for Android Chrome users has resulted in 3 billion fewer scam and spam website notifications per day.
Microsoft shares Insiders preview of Visual Studio 2026
Microsoft has launched its Insiders preview program for Visual Studio 2026, providing insights into what developers can expect from the upcoming release.
One of the main highlights is that the company plans to integrate AI even further into the IDE, describing it as being “woven into the daily rhythms of coding” as opposed to being “bolted on.”
For example, when opening a new codebase, the IDE will suggest the kind of tests that are typically written in the repo and keep docs and comments consistent with the code.
“Code reviews start with clear, actionable insights about correctness, performance, and security – on your machine, before you ever open a pull request. Through it all, you stay in control. The IDE takes the busy-work; you keep the judgment. The result is simple: you move faster, and your code gets better,” Microsoft wrote in a blog post.
Zencoder users can now bring their AI coding tool subscriptions into platform
Zencoder announced an expansion to its platform that lets customers bring popular AI coding tools into Zencoder. New VS Code and JetBrains extensions will allow users to bring their existing ChatGPT, Claude, or Gemini subscription into Zencoder, combining daily limits and enable users to easily switch between models.
“For the first time, developers don’t need to choose between powerful CLIs, IDE integration, or enterprise capabilities,” said Andrew Filev, CEO and Founder of Zencoder. “We’re eliminating tool silos and making AI-assisted development accessible to everyone, from start-ups to enterprise teams alike.”
Microsoft Fabric’s latest update lays foundation for AI
Microsoft announced the latest innovations to Microsoft Fabric at a user conference for the platform, FabCon. Microsoft Fabric is a platform that brings data from multiple sources into one place.
New capabilities were added to OneLake, the unified data lake underlying Fabric, including mirroring capabilities for Oracle and GoogleBig Query, extended support for data agents, and OneLake shortcuts for Azure Blob Storage. Additionally, OneLake now has an integration with Azure AI Search, which will allow users to build more context-aware agents.
And finally, Fabric and Azure AI Foundry are becoming more closely integrated. Fabric provides a way to connect up data and then Azure AI Foundry allows developers to use familiar tools for building and scaling AI applications and agents.
MongoDB MCP Server is now generally available
After a successful public preview, MongoDB announced that its MCP Server is now generally available.
As part of this week’s release, enterprise-grade authentication with OIDC, LDAP, and Kerberos has been added, along with proxy connectivity. There is also now self-hosted remote deployment support so that teams can share deployments and have a centralized configuration.
The MongoDB Server can be downloaded directly or obtained in a bundle with the MongoDB for VS Code extension.
Progress adds AI coding assistance to Telerik and Kendo UI libraries
Progress has announced that it is bringing its AI coding assistants to the Telerik and Kendo UI libraries.
Previously, the company had added AI assistants to Progress Telerik UI for Blazor and Progress KendoReact. According to the company, with today’s release, it now offers AI coding assistance across all major UI component libraries, including ASP.NET Core, WPF, WinForms, .NET MAUI, and Angular.
Progress’ AI coding assistants integrate within developers’ existing IDE workflows and work in AI coding solutions like GitHub Copilot, Claude Code, and Cursor.
They can complete tasks such as generating and configuring components, surfacing relevant API documentation, and resolving component-specific issues, Progress explained.
Redgate’s SQL Prompt updated with new AI features
New features include the ability to use conversational prompts to write SQL code, get explanations of SQL code, get index recommendations to improve performance, and get context-aware instructions for faster query writing in SQL Server Management Studio (SSMS).
These latest features are available to all SQL Prompt or SQL Toolbelt Essentials users, and are opt-in only to give users more control over their use of AI.
“Our priority is giving database professionals the confidence to do their best work,” said Kellyn Gorman, AI Advocate at Redgate. “SQL Prompt has always been trusted because it makes everyday tasks easier, and now we’re extending that with AI in a way that feels supportive rather than disruptive. The new features are designed to work with you: helping to clarify complex queries, improve code quality, and highlight performance opportunities, while keeping you in control of when and how AI is used.”
Mistral announces new connectors, Memories
Mistral announced that its generative AI chat Le Chat now connects with over 20 new connectors, including tools like Asana, Atlassian, Box, Databricks, GitHub, Outlook, Snowflake, Stripe, and Zapier. Users will also now be able to add their own connectors via MCP.
The company also announced a beta for Memories, which allows users to set preferences to get more personalized responses. They can also import their memories from ChatGPT.
Both of these features are available for any Le Chat user, including free users.
OpenAI adds several minor updates to ChatGPT
The company announced that users can now branch off conversations in ChatGPT to explore a specific direction while preserving the direction of the original thread.
Additionally, Projects are now available to free users, and the company has added larger file uploads per project, the option to select colors and icons, and project-only memory controls.
Google announces new open embedding model
EmbeddingGemma is designed for offline, on-device AI, capable of running on less than 200MB of RAM with quantization. It generates embeddings, or numerical representations of text, by “transforming it into a vector of numbers to represent meaning in a high-dimensional space.”
According to Google, embeddings are a crucial part of Retrieval-Augmented Generation, so EmbeddingGemma will enable RAG on mobile devices.
Visa piloting an Acceptance Agent Toolkit
The toolkit will enable non-technical users to build agentic commerce workflows for tasks in Acceptance Invoicing and Pay By Link. For example, a merchant support agent can be given the prompt “create an invoice for $100 for John Doe, due Friday” and it will call the Invoice API, complete details, and send a secure payment link.
Visa also announced its own MCP server to provide an integration layer for agents to access Visa’s capabilities.
“Opening our MCP Server means AI agents can now plug directly into Visa’s infrastructure, access our APIs, and test secure commerce actions. This is an important step in helping AI
developers, partners and clients work with us to build agentic commerce experiences on top of Visa’s payments technology,” the company wrote in an announcement.
Automattic launches experimental AI development tool for WordPress
Telex is a generative AI assistant that can turn natural language prompts into WordPress. For example, a user could ask “I need a reservation block” or “I’d love to add snow to my pages.”
The company’s CEO Matt Mullenweg said “When we think about democratized publishing, like embedded in that, is very core to WordPress’ mission, has been taking things that were difficult to do, that required knowledge of coding or anything else, and … made it accessible to people. Made it accessible in a radically open way, in every language, at low cost, open source — we actually own it and have rights to it,”
Warp releases Warp Code
Warp Code consists of several features for shipping code generated by AI agents. It offers code review capabilities like reviewing open changes, asking for modifications, and line editing code diffs in a dedicated panel. It also has tabbed file viewing, a file tree, and syntax highlighting to improve the editing experience.
“Too often agents write code that almost works, but has subtle issues that end up taking a lot of time to understand, debug, and commit. The solution is not to back away from developing by prompt – instead it’s to improve the prompting workflow so that developers have more comprehension and control. We call this process ‘agent steering’ and our goal with Warp Code is to ship the most ‘steer’-able coding agent around,” the company wrote in an announcement.
Cloudsmith launches ML Model Registry to provide a single source of truth for AI models and datasets
Cloudsmith, providers of an artifact management platform, announced its ML Model Registry, which can act as a single source of truth for all AI models and datasets a company is using.
The registry integrates with the Hugging Face Hub and SDK so that developers can push, pull, and manage models and datasets from Hugging Face and then use Cloudsmith to maintain centralized control, compliance, and visibility.
Once data has been pushed from Hugging Face to Cloudsmith, security and compliance data can be utilized by Enterprise Policy Management so that teams can apply consistent policies to automatically quarantine, block, and approve specific models.
