|
In this newsletter:
- OpenAI releases GPT-5.4 with native computer control and tool search
- Google rolls out Canvas in AI Mode for building apps and docs
- NotebookLM can now turn research notes into fully animated videos
Plus, you’ll find new AI tools and this week’s top AI news headlines!
🚀 OpenAI Releases GPT-5.4 With Native Computer Control and Tool Search
OpenAI launched GPT-5.4 as its most capable model for professional work, now available in ChatGPT, API, and Codex with improved capabilities across knowledge work, coding, and autonomous task execution.
What's new:
- Available now in ChatGPT as GPT-5.4 Thinking, in API as gpt-5.4, and in Codex
- Replaces GPT-5.2 Thinking for ChatGPT Plus, Team, and Pro users (GPT-5.2 remains available for 3 months until June 5, 2026)
- GPT-5.4 Pro is available in ChatGPT and API for maximum performance on complex tasks
- Supports up to 1M tokens of context for planning, executing, and verifying tasks across long horizons
- Most token-efficient reasoning model yet, using significantly fewer tokens than GPT-5.2
Professional work improvements:
- On GDPval testing knowledge work across 44 occupations, matches or exceeds industry professionals in 83% of comparisons vs 70.9% for GPT-5.2
- Spreadsheet modeling tasks score 87.3% mean vs 68.4% for GPT-5.2
- Human raters preferred GPT-5.4 presentations 68% of the time over GPT-5.2 due to stronger aesthetics, greater visual variety, and more effective image generation
- 33% less likely to have false individual claims, 18% less likely to contain any errors in full responses
- ChatGPT for Excel add-in launched today for Enterprise customers
Computer use and vision:
- First general-purpose model with native state-of-the-art computer-use capabilities
- Agents can operate computers and carry out complex workflows across applications
- OSWorld-Verified benchmark shows 75% success rate navigating desktop environment, surpassing human performance at 72.4% and far exceeding GPT-5.2's 47.3%
- WebArena-Verified achieves 67.3% success rate for browser use vs GPT-5.2's 65.4%
- Online-Mind2Web achieves 92.8% success rate using screenshot-based observations alone
- Interprets screenshots of the browser interface and interacts with UI elements through coordinate-based clicking
- Original image input detail level supports full-fidelity perception up to 10.24M total pixels or 6000-pixel maximum dimension
- Improved visual perception on MMMU-Pro reaches 81.2% without tool use vs GPT-5.2's 79.5%
Tool search feature:
- Model receives a lightweight list of available tools instead of all definitions upfront
- Looks up tool definitions only when needed
- MCP-Atlas benchmark with 36 MCP servers enabled showed 47% reduction in total token usage while maintaining the same accuracy
- Enables agents to reliably work with much larger tool ecosystems without sacrificing intelligence
Web search and steerability:
- BrowseComp benchmark shows 82.7% vs GPT-5.2's 65.8%, GPT-5.4 Pro reaches 89.3%
- Stronger at answering questions requiring information from many sources with persistent multi-round searching
- GPT-5.4 Thinking outlines work with preamble for complex queries in ChatGPT
- Users can add instructions or adjust direction mid-response without starting over
- Available now on the web and Android app, coming soon to iOS
- Can think longer on difficult tasks while maintaining awareness of earlier conversation steps
Why it matters:
OpenAI is positioning GPT-5.4 as the model that finally bridges the gap between AI that answers questions and AI that completes entire professional workflows autonomously.
It puts pressure on Anthropic's Claude and Google's Gemini to match the breadth of capabilities in one model rather than requiring users to switch between specialized models for different tasks.
👀Read more about OpenAI GPT-5.4 release!
🔍 Google Rolls Out Canvas in AI Mode for Building Apps and Docs
Google expanded Canvas in AI Mode to all US users in English, allowing them to draft documents, create custom tools, and build shareable apps and games directly from Google Search.
What's new:
- Available now to all users in the US through AI Mode in Google Search
- Previously limited to Google Labs experiments and Gemini subscribers
- Users select the Canvas option from the tool menu (+) while in AI Mode
- Opens Canvas side panel that pulls information from the web and Google's Knowledge Graph
- Can describe an idea and watch as it generates code to transform into a shareable app or game
- Test functionality, toggle to see the underlying code, and refine how the app works by chatting with Gemini
What Canvas can do:
- Draft documents within Google Search
- Build study guides by uploading class notes and other sources
- Turn research reports into web pages, quizzes, or audio overviews (overlaps with NotebookLM)
- Help refine creative writing drafts and get feedback on projects
- Create custom tools and prototypes
For Gemini subscribers:
- Google AI Pro and Ultra subscribers get access to the Gemini 3 model
- 1 million-token context window for more complex projects
How it compares:
- Competes with OpenAI's ChatGPT Canvas and Anthropic's Claude, similar tools
- ChatGPT Canvas triggers automatically based on the query
- Google and Anthropic require more direct interaction
- Both allow help with writing or turning ideas into projects
Why it matters:
Google is using Search's billions of users as a distribution channel to get Canvas in front of people who haven't tried Gemini yet, giving it a massive reach advantage over OpenAI and Anthropic, who rely on users actively seeking out their products.
Making Canvas available in AI Mode instead of requiring a Gemini subscription removes the barrier to trying Google's answer to ChatGPT's code generation and project-building capabilities.
👀 Read more about Google AI Mode upgrades!
🎬 NotebookLM Can Now Turn Research Notes Into Fully Animated Videos
Google's NotebookLM launched Cinematic Video Overviews that generate fully animated videos from research and notes, upgrading from the previous narrated slideshow format.
What's new:
- Available now for Google AI Ultra subscribers over 18 in English
- Uses a combination of Gemini 3, Nano Banana Pro, and Veo 3 AI models
- Gemini determines the best narrative, visual style, format and refines its own work for consistency
- Generates animated visuals based on the content of users' notes
- Users can generate a maximum of 20 cinematic video overviews per day
- Upgrades from the original video overview feature, which could only create narrated slideshows
Why it matters:
NotebookLM is pushing beyond static research summaries into automated video production, positioning itself as a content creation tool rather than just a note-taking app.
The 20-per-day limit and Ultra subscription requirement ($20/month) create barriers for students and researchers who drove NotebookLM's initial viral growth, suggesting Google is testing whether users value AI video generation enough to pay for it before expanding access.
👀 Read more about NotebookLM Video upgrades!
My Latest LinkedIn & X/Twitter Posts:
- How to build slide decks in Claude without switching apps (view post)
- 3 ways you can use AI in your workflows (view post)
- Step-by-step guide for using Perplexity AI (view post)
- Top 9 tips for writing expert-level ChatGPT prompts (view post)
- 15 powerful Perplexity prompts to supercharge your workflow (view post)
In partnership with Kit:
As a creator, your time should be spent doing what you love, not juggling a dozen tools just to run your business.
Kit gives you everything you need in one place:
✅ Build and grow your email list (I use Kit for my newsletter)
✅ Easily monetize with paid newsletters and digital products
✅ Automate your emails with triggers and custom workflows
✅ Track what’s working and optimize with powerful insights
It’s not just email. It’s your entire creator business, simplified.
Join a thriving community of successful creators.
Use Kit for Free — and Start Building Smarter
New AI Tools to Boost Your Productivity:
- Unwrap: Analyzes user feedback to find product insights and trends.
- GPT-5.4: OpenAI's advanced language model for reasoning and task automation.
- Luma: Generates realistic 3D scenes and videos from text.
- LTX-2.3: Generates videos from text with a real-time video model.
- Glaze: Creates desktop apps by chatting with AI.
- Adapt: An AI computer that answers questions and executes workflows across tools.
- GenSong: Generates songs from text prompts.
- Vocova: Converts voice notes into structured text and tasks.
- Monodesk: Workspace for freelance creatives to manage projects and work.
- BusyOcto: An AI marketing platform that creates and executes marketing tasks.
- Stuvio: AI video generator for directing and creating videos.
- GenStore: Generates and launches AI-built online stores.
- CompetitorAnalyzer: Analyzes competitors’ social media strategies and insights.
- Hedy: Conversation coach for speaking clearly and confidently in meetings.
- Bobr: Generates presentation slides in minutes.
This Week's Top AI News Headlines:
- AWS Launches Amazon Connect Health AI Agent Platform to Help Healthcare Providers Automate Patient Support and Administrative Workflows (View Article)
- Luma Launches Creative AI Agents Powered by its New “Unified Intelligence” Models to Automate Video, Image, and 3D Content Creation (View Article)
- Anthropic’s Claude AI Struggles With Surge of New Users as ChatGPT Exodus Overwhelms the AI Chatbot’s Capacity (View Article)
- Microsoft Develops Compact AI Model That Dynamically Decides When to “Think Harder” to Improve Accuracy While Reducing Computing Costs (View Article)
- Netflix Acquires Ben Affleck’s AI Video Startup Interpositive to Advance Machine Learning Tools for Film Restoration and Content Production (View Article)
- Meta Sued Over AI Smart Glasses After Contractors Reportedly Reviewed Private Footage During Content Moderation (View Article)
- AI Shopping Assistants From Companies Like Amazon and Google Are Beginning to Recommend and Automatically Select Products Consumers Buy Online (View Article)
- City Detect, a Computer Vision Startup, Uses AI to Help Cities Automatically Detect Trash, Graffiti, and Infrastructure Problems From Street Camera Feeds (View Article)
- DiligenceSquared, an M&A Research Startup, Launches AI Voice Agents That Automate Due Diligence Calls for Investment Firms (View Article)
- Cursor, an AI Coding Assistant Startup, Rolls Out New Agentic Programming System that Lets Developers Assign Complex Tasks to Autonomous AI Agents (View Article)
- NVIDIA CEO Jensen Huang Says the Company is Pulling Back From Close Ties With OpenAI and Anthropic, Raising Questions About its AI Partnership Strategy (View Article)
- AI in Mental Health is Shifting Therapy From Billable Hour Sessions to Subscription-Based Behavioral Care Blending Human Clinicians and AI Tools (View Article)
Work With Me:
If you enjoy this newsletter, please forward it to your friends and colleagues.
Follow me on LinkedIn and X/Twitter to see my latest posts.
Have a wonderful week!
Andrew Bolis
|