⚡️ OpenAI launches GPT-5.4 Mini & Nano

Good Morning! OpenAI just dropped its latest weapons in the coding wars against Anthropic: GPT-5.4 mini and nano, two insanely fast models built specifically for agentic grunt work. Plus, I’ll show you how to use Canva with Claude AI to generate stunning designs in an instant.

Plus, in today’s AI newsletter:

• Nvidia Restarts AI Chip Manufacturing for China

• Mistral Rolls 4 AI Roles Into One with Small 4

• OpenAI Drops GPT-5.4 Mini & Nano for High-Speed "Vibe Coding"

• How to Use Canva with Claude AI

• 4 new AI tools worth trying

AI MODELS

⚡️ OpenAI Drops GPT-5.4 Mini & Nano for High-Speed "Vibe Coding"

OpenAI just released the smallest and fastest versions of its GPT-5.4 family, designed to be cheap, hyper-fast workhorses for tasks where massive frontier models are just expensive overkill.

• GPT-5.4 mini is over 2x faster than its predecessor (GPT-5 mini) at coding, reasoning, and tool use, making it perfect for debugging or running as a subagent inside Codex.

• GPT-5.4 nano is an even smaller, lighter model designed purely for high-speed API grunt work like data extraction and classification.

• Mini is rolling out to developers via API, Codex, and ChatGPT (accessible via the "Thinking" feature for Free/Go users), while Nano is strictly API-only.

• This launch is a calculated strike at Anthropic’s current dominance in the AI software engineering market, directly challenging the viral success of Claude Code.

As we shift from chatting with AI to managing teams of AI agents, speed and cost become everything. You don't need a massive, expensive frontier model to format a spreadsheet or review a basic pull request. By releasing hyper-efficient models like the 5.4 mini and nano, OpenAI is giving developers the perfect "subagents" to handle repetitive coding grunt work seamlessly in the background.

AI MODELS

⚡ Mistral Rolls 4 AI Roles Into One with Small 4

Mistral is shaking up the open-weight space with Mistral Small 4, a 119-billion parameter Mixture-of-Experts (MoE) model. Instead of maintaining separate models for chat, coding, vision, and reasoning, Mistral baked them all into a single, insanely efficient deployment target.

• Built on a hybrid MoE architecture with 128 experts, activating only 4 per token to keep inference costs low while maintaining a dense knowledge base.

• Features a massive 256k context window and natively handles both text and image inputs.

• Includes a switchable "reasoning effort" parameter, letting you toggle between lightning-fast instant replies and deep, step-by-step problem-solving on a per-request basis.

• Slashes end-to-end completion time by 40% and serves 3x more requests per second compared to Mistral Small 3 in throughput-optimized setups.

• Ships under the permissive Apache 2.0 license with robust deployment support across vLLM and Transformers.

Maintaining separate AI models for different tasks is a massive hardware and resource drain. By unifying chat, reasoning, vision, and coding into a single highly efficient MoE model, Mistral is drastically simplifying how developers build complex agentic workflows. Because it ships with an open license and highly optimized inference, it's a massive win for anyone looking to run powerful, frontier-level AI locally and offline without being tethered to expensive, closed-API ecosystems.

AI NEWS

🇨🇳 Nvidia Restarts AI Chip Manufacturing for China

Nvidia is officially back in business with China. At the company's GTC conference, CEO Jensen Huang announced that Nvidia is restarting production of its H200 chips—designed to comply with strict U.S. export restrictions, after finally securing licenses from both the U.S. and Chinese governments.

• Nvidia had previously halted production of the Hopper-based H200 chip last year due to mounting regulatory tensions between Washington and Beijing.

• The company has now received the green light from the U.S. administration, along with fresh purchase orders from multiple Chinese customers.

• To further capture the highly competitive Chinese AI inference market, Nvidia is also reportedly preparing a China-compliant variant of its newly acquired Groq chips.

• The new Groq LPUs for China are expected to launch as early as May, allowing Nvidia to offer powerful inference capabilities without violating bans on its flagship Vera Rubin systems.

China is a massive market for AI compute, but heavy U.S. export controls and the rise of domestic competitors like Baidu and Huawei have threatened Nvidia's dominance there. By successfully navigating the regulatory minefield to restart H200 manufacturing and aggressively deploying its new Groq technology, Nvidia is proving it will do whatever it takes to maintain its grip on the global AI infrastructure race.

HOW TO AI

🗂️ How to Use Canva with Claude AI to Generate Stunning designs

In this tutorial, you will learn how to connect Claude AI directly to your Canva account, allowing you to generate fully editable logos, social media posts, thumbnails, and presentations instantly using simple text prompts inside the Claude desktop application.

🧰 Who is This For

• Creators needing rapid-fire content generation for Instagram, Facebook, and stories

• YouTubers wanting to cut thumbnail design time from hours to minutes

• Startup Founders and Marketers requiring professional logos, pitch decks, and full brand kits without a design team

• Anyone tired of manual drag-and-drop design who wants fully customizable AI layouts that are editable in Canva

STEP 1: Connect Claude to Canva

First, make sure you are subscribed to a paid Claude plan (Pro, Team, etc.) and have the Claude desktop app installed on your system.

So once you’re ready, open the app, click the “+” icon to start a new chat, and navigate to Connectors. From the menu, find the official Canva connector, click the “+” to add it, and accept the permissions allowing Claude to save designs to your account. Once successful, verify that the Canva connector is enabled in your chat sidebar.

STEP 2: Write Your Design Prompt

Then navigate back to your Claude chat window and describe exactly what you want to create in plain English, but crucially, you must include the word "Canva" in your prompt so Claude knows which tool to use.

For the best results, specify the platform, the product/topic, brand details, and preferred aspect ratio or pixels. For instance, try a prompt like: "Create an eye-catching YouTube thumbnail design in Canva for a video titled 'I tried AI tools for 30 days.' Use a 1280x720 ratio and make it look dramatic."

STEP 3: Review and Select Variations

Claude will process your request and generate several design variations within the chat window. Review these previews carefully. Unlike standard AI image generators that just output a static image, these are editable mockups of actual Canva design files. If you are satisfied, tell Claude which version you prefer by typing, for example, "Go with Option 1." 

STEP 4: Save and Edit in Canva

Claude will confirm that it has saved your selected option directly to your Canva account. Open Canva on the web or your device and navigate to your Projects tab. Your fully generated design will be waiting for you. Open it and begin customizing; unlike static AI results, every element here, fonts, text, shapes, and layouts, is a standard Canva object that you can manually edit, resize, or replace to perfectly match your vision before exporting.

Google expands Personal Intelligence, which lets Gemini tailor its responses by connecting to Gmail and other Google services, from paid users to all US users.

Nvidia is preparing to sell a version of Groq chips to the Chinese market that is expected to be available in May and will not be downgraded.

Micron completes the $1.8B acquisition of Powerchip's DRAM fab in Taiwan, and plans a second fab of similar scale with construction starting by the end of FY26.

Amazon begins three-hour deliveries in ~2K US cities and towns and one-hour deliveries in hundreds of those areas, after 2025 pilots; 90K+ products are eligible.

🤝 Copilot Cowork: Microsoft's agentic AI assistant powered by Claude

🧠 Claude: Anthropic’s AI that can now show interactive diagrams and charts in chat

⚙️ OpenAI Codex: OpenAI’s coding assistant, now with automations and customizable themes

🎥 Replit Animation: Turn text prompts into animated videos

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas