AI Capability Releases: What’s New in May 2026
- claudinenm
- 3 days ago
- 10 min read
Your monthly guide to the latest updates from Claude, ChatGPT, Copilot, and Gemini
May 2026 was defined by two big themes: Google’s I/O developer conference delivered a wave of Gemini announcements that reset expectations for what AI apps can do, and the rest of the industry responded with substantial model and product upgrades of their own. Anthropic shipped Claude Opus 4.8 with meaningful gains in reliability and speed. OpenAI quietly made GPT-5.5 Instant the new default for all ChatGPT users. Microsoft dropped 53 Copilot updates in a single month. And Google unveiled Gemini Omni, Gemini Spark, and a fully redesigned app experience. Here’s what you need to know.
Claude (Anthropic)
Claude Opus 4.8: Sharper Judgment, Faster Execution
Released May 28, Claude Opus 4.8 is described by Anthropic as having “sharper judgement, more honesty about its progress, and the ability to work independently for longer than its predecessors.” The benchmarks back this up:
• Agentic coding score: 64.3% → 69.2%
• Multidisciplinary reasoning with tools: 54.7% → 57.9%
• Knowledge work score: 1,753 → 1,890
• Fast mode is now roughly 2.5× quicker than Opus 4.7
Pricing is unchanged from Opus 4.7, making this a straight quality upgrade for existing users. The speed improvement in fast mode is particularly significant for agentic workflows where Claude is executing many sub-tasks in sequence.
Dynamic Workflows: Tackling Bigger Problems in Claude Code
Anthropic launched Dynamic Workflows in research preview, a new capability in Claude Code that enables it to take on significantly larger and more complex engineering challenges. Rather than tackling a single well-defined task, Dynamic Workflows allow Claude to break down, plan, and execute across broader problem spaces — moving closer to the kind of open-ended engineering work that typically requires a human to orchestrate.
Messages API: Mid-Task Instruction Updates
A quiet but developer-significant update: the Messages API now accepts system entries inside the messages array. This allows developers to update Claude’s instructions mid-task without breaking the prompt cache or having to route the update through a user turn. For complex agentic pipelines where instructions need to evolve as a task progresses, this is a meaningful quality-of-life improvement.
Managed Agents “Dreaming”: Smarter Long-Term Memory
Anthropic introduced “Dreaming” for its Managed Agents API — a research preview feature that consolidates an agent’s persistent memory between sessions by merging duplicates and removing stale entries. Think of it as a memory housekeeping pass that runs between conversations, keeping the agent’s long-term context clean and relevant rather than letting it accumulate noise over time.
ChatGPT (OpenAI)
GPT-5.5 Instant Is Now the Default for Everyone
On May 5, OpenAI replaced GPT-5.3 Instant with GPT-5.5 Instant as the default model for all ChatGPT users. The upgrade is automatic — no settings change required. GPT-5.5 Instant delivers measurably reduced hallucination rates in high-stakes domains (law, medicine, finance) while maintaining the low-latency feel users expect from the Instant tier. For the vast majority of users who never change their model setting, this is a meaningful quality improvement that simply appears one day.
ChatGPT Comes to Excel and Google Sheets
Also on May 5, OpenAI launched ChatGPT for Excel and Google Sheets globally. ChatGPT now lives in the sidebar of both spreadsheet apps, letting users build, update, and understand data without leaving their spreadsheet. You can ask it to write formulas, explain what a formula does, clean messy data, or generate new columns based on existing ones — all in plain English, directly inside the tool you’re already using.
Personalized Search Across Past Conversations, Files, and Gmail
GPT-5.5 Instant gained a new capability for Plus and Pro users: the ability to search across past conversations, uploaded files, and Gmail to give more personalized, context-aware answers. Ask “What was the name of that vendor we discussed last month?” and ChatGPT can actually look it up. This expands ChatGPT’s memory from “what you’ve told it” to “what it can find,” a meaningful distinction for power users with rich conversation histories.
Memory Controls: Delete, Edit, and Turn Off
ChatGPT users can now manage their memory directly from the memory summary page. Available to all users on the web (rolling out to mobile), the new controls let you delete specific memories, select “Delete and turn off memory” entirely, or edit what’s been saved. This gives users meaningful agency over what ChatGPT knows about them — addressing one of the most common concerns about AI personalization.
Codex: Goal Mode, Conversation Search, and Richer MCP Support
OpenAI’s Codex CLI received a series of updates throughout May, including Goal Mode enabled by default (letting Codex pursue a defined outcome across multiple steps), conversation search, and significantly richer MCP (Model Context Protocol) support. For developers using Codex in their workflows, MCP support expansion means it can now connect to a broader range of tools and data sources natively.
GPT-5.5-Cyber: A Specialized Security Model
On May 7, OpenAI announced GPT-5.5-Cyber under their Trusted Access for Cyber program — a limited-preview cybersecurity variant of GPT-5.5. This follows Anthropic’s Claude Mythos preview from April, suggesting the two leading AI labs are both moving seriously into the security vertical.
Microsoft Copilot
53 Updates in One Month
Microsoft’s May release for Microsoft 365 Copilot was its most extensive yet: 53 individual updates across the product suite. Here are the highlights that matter most.
Computer-Using Agents: Now Generally Available
After months in preview, computer-using agents are now generally available in Copilot Studio. Organizations can build agents that interact directly with websites and desktop applications through their user interfaces — clicking, typing, and navigating like a human would. This unlocks automation for processes that previously required brittle RPA scripts or manual work, particularly in legacy systems with no API.
Redesigned Workflows in Copilot Studio
Copilot Studio’s workflow builder was overhauled with a new unified visual canvas. Users can now see how actions, decisions, and AI-powered steps connect end-to-end in one place, rather than stitching together disconnected tools. The redesign makes it significantly easier to build, debug, and modify complex agentic automations.
Claude Opus 4.8 and GPT-5.5 Instant Now Available in Copilot
Microsoft expanded model choice in May in two directions at once: Anthropic’s Claude Opus 4.8 is now available in Microsoft 365 Copilot for complex, multi-step tasks and long-running workflows, and OpenAI’s GPT-5.5 Instant is now available in both M365 Copilot and Copilot Studio for faster, more concise everyday responses. Having best-in-class models from multiple providers available within a single product is a meaningful enterprise advantage.
Copilot Notebooks: Teams Meetings and Excel Generation
Copilot Notebooks received two significant updates:
• Teams meeting integration: Transcripts, notes, chats, and shared content from individual Teams meetings can now be added as knowledge sources, grounding your notebooks in actual meeting context
• Excel agent integration: Notebooks can now generate tailored Excel spreadsheets directly from notebook content, with the Excel agent asking clarification questions to customize the output
Federated Copilot Connectors via MCP
Microsoft launched federated connectors for Canva, HubSpot, Linear, LSEG, Moody’s, and Notion via MCP. This means users can bring data and actions from these external tools directly into Copilot without leaving Microsoft 365 — a significant expansion of Copilot’s reach into the broader tool ecosystem.
Plan Mode and Python in Copilot for Excel
Copilot for Excel gained Plan Mode, which shows users a step-by-step plan before executing complex data operations, and native Python support — enabling advanced data analysis, custom visualizations, and statistical modeling directly in Excel without leaving the app.
Gemini (Google)
Google I/O 2026: The Month’s Biggest AI Event
Google’s annual developer conference on May 19 was dominated by Gemini announcements. Nearly every major update below came out of or in the weeks around I/O 2026.
Gemini 3.5 Flash: Frontier Intelligence Meets Agentic Action
Google’s new flagship efficient model, Gemini 3.5 Flash, outperforms Gemini 3.1 Pro on key benchmarks while costing less than half as much. It leads in multimodal understanding and is specifically optimized for long-horizon agentic tasks — the kind of multi-step workflows where the model needs to plan, use tools, and follow through over an extended period. Gemini 3.5 Pro is in testing and expected next month.
Gemini Omni: Generate Any Output from Any Input
Gemini Omni is Google’s most ambitious model yet — a multimodal model that can take any combination of text, audio, image, and video as input and produce any output type, with video generation as its flagship capability. Gemini Omni Flash is rolling out to developers and enterprise customers via the Gemini API and Agent Platform API in the coming weeks. For content creators and developers building multimedia experiences, this opens entirely new possibilities.
Gemini Spark: Your 24/7 Personal AI Agent
Google introduced Gemini Spark, a new personal AI agent designed to help users navigate their digital lives proactively — not just when you ask, but continuously. Spark represents Google’s most direct answer to the concept of an always-on AI assistant, one that can take initiative and act on your behalf rather than waiting to be prompted.
Daily Brief: A Personalized Morning Digest
The Gemini app now includes a Daily Brief — a personalized digest designed to be your first stop each morning. It pulls together information from your inbox, calendar, and most important tasks, then organizes them into a clear overview with prioritized to-dos and suggested next steps. For users already relying on Gemini’s Personal Intelligence integration with Google apps, this brings that context into a structured daily format.
Neural Expressive: A Redesigned App Experience
The Gemini app launched a new design language called “Neural Expressive” — featuring fluid animations, vibrant colors, updated typography, and haptic feedback. It’s a significant visual refresh from the previous interface, and signals Google’s intent to position Gemini as a consumer product, not just a developer tool.
Google Antigravity: An Agent-First Development Platform
For developers, Google advanced Antigravity, their agent-first development platform designed to move beyond AI tools that help you write to agents that help you act. Antigravity positions Google’s developer ecosystem around agentic AI as a first-class primitive — a foundational shift in how Google wants developers to think about building with AI.
What Does This Mean for Users?
May was a month of practical upgrades. Less “here’s what AI will eventually be able to do” and more “here’s what you can do today that you couldn’t do last month.” Here’s the user-facing impact of everything above.
Your spreadsheet app just got a capable AI assistant built in
The global launch of ChatGPT for Excel and Google Sheets is one of the most practically significant releases of the month, precisely because it meets users where they already are. Most knowledge workers spend significant time in spreadsheets. Now, instead of switching to a separate AI tool, writing a formula, or searching for documentation, you can describe what you want in plain English — and get it. For finance, operations, and data teams, this alone is worth the subscription cost.
AI agents can now use your actual software, not just APIs
The general availability of computer-using agents in Copilot Studio is a watershed moment for enterprise automation. Until now, AI automation required APIs — meaning only modern, well-integrated software could be automated. Computer-using agents change that: they can navigate legacy systems, internal tools, and any website just like a human employee would. If you’ve ever thought “I wish I could automate this but it doesn’t have an API,” May 2026 is your answer.
The best AI models are now available inside Microsoft 365
Having both Claude Opus 4.8 and GPT-5.5 Instant available within Microsoft 365 Copilot means enterprise users no longer have to choose a single AI provider or maintain separate subscriptions to access best-in-class models. You can route complex analytical work to Opus 4.8 and quick everyday tasks to GPT-5.5 Instant, all within the same Copilot interface. For organizations already standardized on Microsoft 365, this is a significant value expansion.
AI is starting to anticipate your day, not just respond to it
Gemini’s Daily Brief and Gemini Spark together represent a shift in the AI paradigm from reactive to proactive. Instead of opening an AI app to ask a question, these features surface what you need before you ask — your schedule, your priorities, your most important emails, organized and ready. For users with packed mornings and information-dense jobs, this kind of proactive briefing can meaningfully reduce the cognitive overhead of starting a workday.
Video generation is now a practical tool, not just a demo
Gemini Omni’s ability to generate dynamic video from any combination of text, images, audio, and video moves generative video from novelty to utility. For marketers creating product content, educators building explainer videos, or professionals who need visual communication quickly, Omni makes video generation accessible at the same level of effort as writing a prompt. The Omni Flash variant rolling out via API means developers can start building products on top of it now.
AI memory is becoming something you can actually manage
ChatGPT’s new memory controls — delete, edit, turn off — address a real anxiety that many users have about AI personalization: what does it know about me, and can I fix it if it’s wrong? The ability to see and edit your memory summary directly transforms AI memory from a black box into a visible, manageable profile. This trust-building feature matters especially for users who’ve been hesitant to let AI systems accumulate context about their work and preferences.
Claude is becoming more reliable for long-running autonomous tasks
The combination of Claude Opus 4.8’s improved agentic coding scores, Dynamic Workflows in Claude Code, and the mid-task instruction update in the Messages API points at a consistent theme: Claude is being systematically improved for long autonomous runs. If you’ve used Claude for agentic tasks before and found it needed too much hand-holding, May’s updates are worth a re-evaluation. The model is more capable, faster, and easier to steer mid-task than it was a month ago.
Bottom Line
May 2026 was a month where the gap between “AI as a tool you use” and “AI as a colleague you work with” narrowed perceptibly. Spreadsheets got AI sidebars. Agents learned to use real software. Morning briefings became personalized. Models got faster, more honest, and more capable of sustaining long tasks without human correction. The aggregate effect is an AI landscape that feels meaningfully more useful in everyday work than it did 30 days ago.
Want this roundup delivered to your inbox every month? Hit Subscribe. And if you found a capability we missed, drop a comment — the AI space moves fast and we want to keep this as comprehensive as possible.
Sources
Comments