OpenAI unveils o3 and o4-mini

ALSO: How to create visual effects with AI

Welcome back, Superhuman. Imagine hiring a genius and then taking away their internet access, calculator, and reading glasses all at once. That’s basically what we were doing with LLMs…until now. With o3 and o4-mini, OpenAI just showed how tool use can push reasoning models to new heights.

Today’s Insights

  • OpenAI’s surprise coding tool, Microsoft’s agents, and Claude’s voice mode

  • Everything to know about o3 and o4-mini

  • Tutorial: How to create stunning visual effects in minutes with AI

  • 5 new AI tools to boost your productivity

  • News, memes, what’s trending on socials, and more

TODAY IN AI

A video shows OpenAI’s new Codex CLI coding tool in action. Source: OpenAI

1. OpenAI unveils o3 and o4-mini, plus a special bonus: For the first time, OpenAI’s reasoning models can tap into the same tools you use to solve problems. But first, let’s talk about a big surprise: Codex CLI. This open-source agent can connect to your computer and help with coding tasks. In a demo, a researcher took a screenshot of an app someone else had generated and then asked Codex to recreate it locally. He even added custom instructions, like making the app compatible with a webcam. The real game-changer is that it’s open-source. (Keep reading for more on o3 and o4-mini.)

2. Microsoft’s Copilot Studio can now take control of your computer: The tech giant’s AI developer platform just learned how to navigate the web on its own. You can design agents to perform market research, process invoices, and complete data entry tasks autonomously. As The Verge points out, the consumer version of Copilot got a similar update a few weeks ago, but this one seems even more expansive, giving the platform complete freedom to visit any webpage, not just specific sites.

3. Anthropic set to release long-awaited voice mode: OpenAI’s Advanced Voice mode could soon face some stiff competition. Meet Airy, Mellow, and Buttery — three new voice-based personalities that Anthropic could bring to its Claude chatbot as early as this month. The startup might be the last big player in the AI space without a voice assistant, so the update could be just what it needs to win back users who’ve drifted away in recent months.

PRESENTED BY TURING

Is your GenAI strategy working? Find out in 5 minutes

Need help deploying GenAI or enhancing your current strategy?

It only takes 5 minutes. With Turing’s no-cost assessment, you’ll discover where you are on your GenAI journey (and what to do next).

Perfect for startups, enterprises, and everyone in between:

FROM THE FRONTIER

OpenAI’s o3 and o4-mini combine reasoning and tool use for the first time

OpenAI unveiled o3 and o4-mini on Wednesday, paving the way for GPT-5. Source: OpenAI

A lot of human progress, from agriculture to cities, comes down to our innovative use of tools. They help us preserve knowledge so we don’t waste time on problems that other people have already solved.

That explains why it’s such a big deal that OpenAI’s new models, o3 and o4-mini, just gained the ability to use all of the tools already embedded in ChatGPT. Instead of just reasoning through a problem on their own, o3 and o4-mini can tap into things like web browsing, image understanding, and the Python coding language to approach the same prompt from multiple angles.

The evidence: The models are now considered state-of-the-art across nearly every math and science category. o4-mini got a near-perfect score on the AIME 2025 Competition Math benchmark. And on the Codeforces Competition Code eval, o3 and o4-mini (with tools) scored among the top 200 contestants in the world.

The models can also ‘think’ with images for the first time: For example, if you upload an intricate diagram and ask a question about it, ChatGPT will zoom in on relevant figures and solve equations related to them all on its own.

The result: “These are the first models where top scientists tell us they produce legitimately good and useful, novel ideas,” OpenAI President Greg Brockman said during Wednesday’s livestream. Builder McKay Wrigley added that if OpenAI can just give o3 a little more freedom to browse the internet, you’d basically have “a scalable remote worker right there.” GPT-5 is really starting to come into picture.

THE AI ACADEMY

How to create stunning visual effects in minutes with AI

  • Go to Higgsfield AI and sign up.

  • Choose the motion controls of your choice from the list to add motion to your image.

  • Upload your image and enter your prompt describing the scene you imagined with details.

Sample Prompt: skyscraper in background explodes, while a woman looks at a viewer with a shocked face, blast wave blows her hair

  • Click on ‘Generate’ and wait for a few seconds.

  • You’ll get stunning, high-quality VFX ready for you without much effort.

PRESENTED BY INNOVATING WITH AI

The Tools, Templates & Playbook for Your AI Consultancy

Inside The AI Consultancy Project, you'll find dozens of templates and toolkits to quickly grow an AI consulting business – even if you're not a techie.

Join 700+ students and coaches who've been featured in Wired, Entertainment Weekly, and TechCrunch.

AI & TECH NEWS

Everything else you need to know today

OpenAI could soon acquire AI coding assistant Windsurf. Source: Windsurf

🤝 Coding Conquest: OpenAI is reportedly in talks to buy AI coding startup Windsurf (formerly known as Codeium) for $3B, in what would be the ChatGPT-maker’s largest acquisition yet.

🗂️ Easy Edits: xAI’s Grok can now interact with your Google Drive files. The platform is also getting a canvas-like feature that lets you quickly edit text and code inside the app.

📉 Market Woes: The US is cracking down on Nvidia’s exports to China with new restrictions on its H20 chips (the less powerful ones it designed specifically for the Chinese market). The revelations sent tech stocks tumbling Wednesday, costing Nvidia an estimated $5.5B.

🔬 Scaled-Up Science: Biotech startup Profluent says it has discovered a scaling law for AI-powered protein design tools — meaning it’s proven for the first time that more data leads to better results, even for biology-focused LLMs.

🛍️ Shopping Spree: Immersive AI startup Infinite Reality just acquired Touchcast — whose Mentorverse can generate life-like avatars trained on your company’s data — for $500M. That comes only a few weeks after it scooped up retro file-sharing platform Napster.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

 Nily: An AI assistant with 20+ AI models, LLM response comparisons, and mixture AI for delivering the most optimal answer.

 SpreadSimple: Turn your Google Sheets into a website with no code.

 ScrapeGraphAI: Transform any website into clean, organized data for AI agents and data analytics.

 Extrovert: Track your customers' posts, spot key conversations, and help your team build trust with thoughtful comments.

 Plus AI: Create stunning presentations and edit slides with AI.

 🎁 Want more? Check out our Top 125 AI Tools

* indicates a promoted tool, if any

PROMPT OF THE DAY

Break down any job posting

Prompt: Act as a seasoned hiring manager in [insert industry or role]. Analyze the following job description and identify the top 3 skills or traits the employer values most, even if they’re not explicitly stated. 
Tell me what kind of problems this role is likely responsible for solving and how I can tailor my resume and cover letter to align with those priorities. Highlight which keywords I should include to pass ATS filters and suggest the types of interview questions I might be asked based on the description. Also, share what qualities or signals would help me stand out from other applicants. Finally, craft one strong sentence I can include in my cover letter or outreach email that shows I clearly understand what they’re looking for. 
Here’s the job description: [paste job description].

🤓 Want more? Check out our Top 1,000 Prompts

SOCIAL SIGNALS

What’s trending on socials today

👨‍💻 Never Too Young: Builder Tashfeen Suleman said his 9-year-old son had so much fun visiting an escape room that he created his own puzzle game inspired by the experience with the help of AI coding assistant, Replit.

🤖 Less is More: Programmer Thorsten Ball shared an in-depth guide for how to build an agent that can edit code on your behalf. He claims you can do it with less than 400 lines of code.

🚠 Going Downhill: What if OpenAI’s GPT-4.1 was actually a downgrade from GPT-4.5, as its name suggests, with even worse models to come? That’s one researcher’s tongue-in-cheek prediction.

👀 o3 Reaction: Professor and researcher Derya Unutmaz says that OpenAI’s new o3 model outperforms him in his domain of biomedical science and claims it is as good as the top 10% of professor-level top scientists. Writer Dan Shipper also explained why o3 is so special.

AI-GENERATED IMAGES

Polly Pocket toy

Source: @hc_dsn on X

ChatGPT Prompt: Create a realistic portrait-ratio photo of a Polly Pocket-style dollhouse: a heart-shaped compact with shiny 80s/90s toy textures. The theme is an [ENTER THEME]. Outside: coffee stains, doodles, and a sticker-covered laptop. Inside: pale yellow and baby blue with muted tones. Rooms include a cluttered dual-monitor desk, nap nook, plant zone, and design-crisis lounge. Accessories: [ADD ACCESSORIES]. The doll has a messy bun, eyebags, and retro proportions to match the chaotic, cute scene.

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 1M+ readers and 2M+ followers on socials working at the world’s leading startups and enterprises. Companies like Amazon, Hubspot, and Salesforce feature their products in Superhuman. You can learn more about partnering with us here.

🧞 Your wish is my command

What did you think of today's email?

Your feedback helps me create better emails for you!

Login or Subscribe to participate in polls.

Got more feedback or just want to get in touch? Reply to this email and we’ll get back to you.

Thanks for reading.

Until next time!

Zain & the Superhuman AI team