Anthropic unveiled Claude Fable 5 and Claude Mythos 5, its next generation of models for the hardest coding and knowledge-work problems. Fable 5 is the first Mythos-class model made safe for general use, and it went generally available in GitHub Copilot the same day.
VS Code 1.124 makes agents smarter at determining task completion with improved Autopilot enabled by default, adds background sessions for parallel request composition, and introduces session navigation plus browser history search. The Agents window companion app continues refinement with keyboard-driven session management and cross-machine iteration support.
GitHub Copilot Chat can now read the status of your in-progress agent sessions and query past ones. Two new tools let chat pull cloud-agent session logs and search prior sessions by topic, title, or recency.
Google's DiffusionGemma is an experimental Gemma 4 model that generates text in parallel using diffusion instead of token-by-token autoregression, delivering up to 4x faster inference with built-in self-correction while running on consumer GPUs.
GitHub Copilot CLI now ships an experimental /security-review slash command that scans your local code changes for high-impact vulnerabilities and returns actionable fixes without leaving the terminal.
OpenAI and Oracle are partnering so OCI customers can apply eligible Oracle Universal Credits toward OpenAI frontier models and Codex, giving enterprises access through purchasing workflows they already trust.
Anthropic's new Mythos-class model, Claude Fable 5, is now selectable in the GitHub Copilot model picker across VS Code, Visual Studio, Copilot CLI, the cloud agent, and more. It targets long-horizon, autonomous coding and is available to Pro+, Max, Business, and Enterprise plans.
Apple developers can now securely call cloud-hosted Gemini models using the Foundation Models framework and access Gemini directly in Xcode. The integration leverages Firebase AI Logic for production-grade service without requiring a separate backend server, and includes self-serve keys plus enterprise options for organizational control.
Anthropic announced at Code w/ Claude that Claude Managed Agents can now operate in customer-controlled sandboxes and connect to private MCP servers, keeping tool execution and services within enterprise boundaries. The platform also added nested sub-agents working in parallel, outcomes-based grading, and webhooks for long-running tasks.
Microsoft Agent Framework hits general availability, converging AutoGen and Semantic Kernel into a single open-source SDK for .NET and Python. The release includes agent harness for execution control, hosted agents on Azure, and native support for MCP integrations and multi-step workflows.
GitHub brought enterprise-managed plugins to VS Code on June 5, extending the May 6 Copilot CLI preview to the editor. Admins can now push agents, skills, hooks, and MCP servers to every developer through a single policy file, with VS Code 1.122 also removing the GitHub OAuth requirement for BYOK in air-gapped environments.
GitHub Copilot now supports one-million-token context windows and configurable reasoning levels in VS Code, Copilot CLI, and the Copilot app, letting developers trade off speed, depth, and AI credit consumption per task.
VS Code 1.123 introduces automatic chat session syncing across machines via GitHub accounts, support for 1 million-token context windows for Anthropic and OpenAI models, and the ability to open multiple agent sessions side-by-side. The update includes a research agent preview, chronicle commands for querying past work, and enhanced integrated browser tools.
GitHub ships major updates to Copilot CLI including generally available rubber duck mode, prompt scheduling, voice input, and an experimental terminal interface with tabs for issues, pull requests, and gists. The CLI also gains a redesigned UI with theme-aware colors and responsive components that adapt to narrow terminals.
The /chronicle command turns your Copilot session history into standup summaries, personalized tips, and custom instructions. It now spans the cloud agent, code review, the Copilot app, VS Code, and JetBrains, with local sessions synced to your GitHub account.
Microsoft announces MAI-Code-1-Flash, its first proprietary coding model built for GitHub Copilot and VS Code with 5 billion active parameters at lower cost than Haiku, plus MAI-Thinking-1 reasoning model matching Sonnet 4.6 in blind evaluations. The models signal Microsoft's push toward independence from OpenAI and Anthropic with clean-data training and no third-party distillation.
GitHub added Gemini models as a selectable option across the Copilot CLI, the cloud agent, and the new Copilot app, broadening multi-model choice in agentic Copilot surfaces.
GitHub Copilot can now run inside isolated sandboxes, both locally on your machine and in ephemeral GitHub-hosted cloud environments. The release gives agentic workflows a secure execution layer with consistent identity, governance, and policy controls built in.
OpenAI introduced six role-specific Codex plugins, a preview of shareable hosted Sites, and annotations for in-place refinement of documents, slides, and spreadsheets. Codex now serves more than 5 million weekly users, with non-developers making up about 20 percent and growing 3x faster than developers.
GitHub unveils the Copilot app at Microsoft Build 2026, described as an agent-native desktop experience that surfaces tools and updates for working across VS Code, CLI, and mobile. Remote control for Copilot sessions is now generally available on github.com and GitHub Mobile, enabling developers to kick off work in one surface and finish on another.
The Copilot SDK is generally available, giving developers programmatic access to Copilot's agent runtime - planning, tool invocation, file edits, streaming, and multi-turn sessions - so they can embed it in their own apps and tools without building an orchestration layer.
Cognition rebranded Windsurf as Devin Desktop on June 2, 2026, unifying its agentic IDE with Devin cloud agents under one brand. The release introduces Agent Command Center for managing local and cloud agents from a single Kanban view, plus support for the open Agent Client Protocol.
GitHub updated Copilot billing and plans so each interaction consumes AI credits based on token cost and the model you choose. VS Code surfaces per-model costs and credit consumption in the model picker and status dashboard.
OpenAI frontier models and Codex are generally available on AWS through Amazon Bedrock, opening a path for millions of AWS customers to build with OpenAI using their existing security, compliance, and governance workflows in both Commercial and GovCloud regions.
Google launches Antigravity 2.0 with a new CLI, managed agents in isolated Linux environments via the Gemini API, and WebMCP - an experimental open standard enabling browser-based AI agents to execute structured tasks. The release also brings native Android app generation to AI Studio and terminal sandboxing with Git policies.
Anthropic rolled out Claude Code as a standalone agentic coding system, not just a tool. The company announced it now writes most of Anthropic's own code and added a real-time security-guidance plugin that flags SQL injection, XSS, and hardcoded secrets inline during coding sessions.
OpenAI released Codex version 26.527 on May 29, bringing Computer Use capabilities to Windows for the first time. The update lets Codex operate Windows desktop apps by seeing, clicking, and typing in the foreground, plus adds mobile remote control from ChatGPT on iOS and Android so developers can start, review, and steer coding tasks on the go while work continues on their Windows machine.
Anthropic released Claude Opus 4.8 on May 28, just 41 days after Opus 4.7. The model ships with Dynamic Workflows, a research preview feature that lets Claude Code orchestrate hundreds of parallel subagents in a single session, executing codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge. Opus 4.8 is also four times less likely than its predecessor to let flawed code pass unremarked.
VS Code 1.122 ships air-gapped BYOK support for local models, browser device emulation for testing responsive web apps, and enhanced agent session management. The release lets developers use their own language models completely offline with Ollama and other providers, no GitHub sign-in required.
OpenAI's engineering team published a deep-dive on building self-improving tax agents powered by Codex - a case study in evaluation-driven agent development and continuous fine-tuning loops.
Cursor published a case study showing how Faire doubled pull-request throughput after adopting Cursor Cloud Agents. Hard productivity data from a recognizable retail-tech company.
Microsoft is cancelling most direct Claude Code licenses after the tool spread beyond engineering to non-technical roles and burned through budgets faster than expected. The move signals broader industry repricing of usage-based AI tooling as token costs exceed predictions.
Gartner placed OpenAI in the Leaders quadrant of its 2026 Magic Quadrant for Agentic Coding. The recognition formalizes Codex's positioning against GitHub Copilot, Cursor, and Anthropic's Claude Code in the enterprise.
GitHub has open-sourced the Copilot extension for Eclipse, opening up community contributions for one of the most-used enterprise IDEs and signaling continued commitment to multi-IDE parity.
Copilot in VS Code now auto-routes requests to the best model based on the task at hand - chat, edits, or agent mode. For Pro+ and Business users, this can lower cost and latency without manual model picking.
VS Code 1.121 shipped May 20 with remote agent sessions, allowing developers to monitor and control agent work on a remote machine from the Agents window. The release adds Mermaid and HTML previews built-in, configurable model selection for lightweight tasks like commit messages, and terminal tool optimizations to reduce token usage for agent workflows.
KPMG announced a strategic alliance integrating Claude across its core business and 276,000-person global workforce. The deal is one of the largest single-vendor enterprise AI rollouts to date.
Cursor now connects directly to Jira, letting agents pick up issues, propose changes, and report status without leaving the editor. It is a step toward Cursor's "self-driving codebase" vision where agents own the issue-to-PR loop.
Google releases Gemini 3.5 Flash, outperforming Gemini 3.1 Pro on coding and agentic benchmarks (76.2% Terminal-Bench 2.1, 83.6% MCP Atlas) while delivering 4x faster output token generation. Available today in Antigravity, Gemini API, and Google Search AI Mode.
Google unveils Gemini 3.5 Flash with frontier intelligence at 4x speed, Antigravity 2.0 desktop application for agent-first development, and Managed Agents in the Gemini API powered by the Antigravity agent harness. The update includes native Android vibe coding in AI Studio, WebMCP origin trial in Chrome 149, and Modern Web Guidance for 100+ use cases.
Google ships Antigravity 2.0 desktop app, Managed Agents API for spinning up agents in isolated Linux environments with a single API call, and native Android app generation in AI Studio. Powered by Gemini 3.5 Flash, the Managed Agents feature uses the same Antigravity agent harness that powers Google's internal tools.
Gemini 3.5 Flash is now generally available across GitHub Copilot, giving developers a fast, cost-efficient option for completions and chat that complements the heavier OpenAI and Anthropic models in the picker.
Google launched Antigravity 2.0 desktop app, Gemini 3.5 Flash model, and Managed Agents in the Gemini API at I/O 2026 on May 19. Developers can now spin up agents with a single API call using the Antigravity agent harness in Google-hosted sandboxes, with native Android vibe coding in AI Studio and a new $100 AI Ultra plan.
Google launched Antigravity 2.0 at I/O 2026 with Gemini 3.5 Flash as the default model, a new standalone CLI for terminal workflows, and a Managed Agents API for enterprise deployments. The release adds specialized subagents for parallel tasks and cross-platform sandboxing.
Google I/O 2026 launched Gemini 3.5 Flash on May 19, combining frontier intelligence with agentic task execution at 4x faster output than other frontier models. The company upgraded Antigravity to version 2.0 with subagent orchestration, built-in terminal sandboxing, and the new Antigravity CLI for developer productivity gains.
Anthropic acquired Stainless, the company behind one of the most widely used SDK-generation platforms powering API clients for OpenAI, Anthropic, Stripe, and others. The deal puts a critical piece of the developer-API supply chain inside Anthropic.
Cursor unveiled Composer 2.5, a substantial improvement over Composer 2 on long-horizon agentic tasks. It is the company's bid to stay ahead in the autonomous-coding race against Copilot cloud agent and Claude Code.
OpenAI partnered with Dell Technologies to bring Codex to hybrid and on-premises enterprise environments, targeting regulated industries that have been unable to use cloud-only AI coding tools.
The Copilot Spaces API has reached general availability, letting teams programmatically create, share, and manage curated context collections that scope what Copilot can see across repos and docs.
GitHub promoted remote control of Copilot CLI sessions to general availability across mobile, web, and VS Code. Developers can now resume long-running CLI agent tasks from anywhere.