This week both Anthropic and OpenAI dropped interactive visual tools within 48 hours of each other. Same concept, wildly different approaches. And that’s just the appetizer.
The creator behind the breakdown is Matt Wolfe, who stress-tested both platforms side by side so the rest of us don’t have to. The results are genuinely surprising.
🔍 Claude vs ChatGPT: Visual Generation Showdown
Here’s the twist: Claude generates every visualization from scratch. ChatGPT pulls from a library of pre-built templates. Claude took about 90 seconds to render a compound interest calculator with working sliders. ChatGPT did it almost instantly, but when asked for anything outside its template library, it flat-out said it couldn’t do it. Two very different philosophies, both available on free plans.
The Claude version nailed compound interest and AI timeline visualizations but fell apart on maps and neural network diagrams. ChatGPT’s pre-built visuals looked polished (animated gas molecules, Pythagorean theorem sliders) but you’re locked into supported categories only.
📋 Other Major Releases This Week
- Canva Magic Layers splits any image into movable layers automatically. Upload an AI-generated image, click Magic Layers, and suddenly the background, subject, and individual objects become independent pieces you can rearrange. Works with real photos too. Available on all Canva plans.
- Perplexity Computer expanded to all paid plans. It now runs on dedicated Mac Minis that work 24/7 as your digital proxy. The expert tested it by asking for an interactive AI company map, and it delivered something significantly better than Claude’s attempt, though it burned about 230 credits and took 5 minutes.
- Andrej Karpathy open-sourced Autoresearch, a system that lets AI models optimize themselves overnight. You give it a training setup, it runs experiments autonomously, keeps improvements, discards failures, and repeats. You wake up to a log of experiments and a better model. The code is on GitHub right now.
- Meta acquired the team behind Moltbook, the social network built entirely for AI agents. The expert floats two theories: Meta wants to cut creators out of the content loop entirely, or they’re positioning to influence AI agents that will increasingly make purchasing decisions on behalf of humans. Both theories point the same direction.
🛠️ Quick Hits Worth Knowing
- Nvidia released Nemotron 3 Super, a 120B parameter open-weight model
- Google shipped Gemini Embedding 2, their first natively multimodal embedding model
- ChatGPT for Excel launched with a sidebar that reads and writes spreadsheet data
- Claude Code got scheduled tasks (daily code reviews, weekly audits) and an agent-based code review system
- Microsoft Copilot Health now aggregates medical records and wearable data
- Google Maps lets you have full conversations about your surroundings and route planning
- Figure showed off a room-tidying humanoid robot (though the room was already pretty clean)
💡 Pro Tip for Choosing Tools
If you need a specific educational visualization, start with ChatGPT’s pre-built library for polished results. If your concept isn’t covered there, switch to Claude and let it build one from scratch. Different tools, different strengths.
The full video walks through live demos of every single feature mentioned here. Definitely worth watching if you want to see these tools in action before deciding which ones fit your workflow.