🤯 OpenAI’s Latest Drops!

Okay, what is OpenAI on lately?! Because these guys have been on a nonstop release spree.

First came GPT-4.1. And its sibling, and now?

They’ve casually dropped the long-anticipated o3 and o4-mini — two new reasoning models that are basically the Sherlock Holmes of AI: sharp and methodical, processes answers with care, think things through, and adds the perfect touch of extra insight when you need it most.

Here’s the scoop in plain English (and good vibes, obviously):

• o3 is the brainiac of the bunch. OpenAI says it’s their most advanced reasoning model ever. It crushes tests in math, coding, science, and even visual understanding — like, it can literally make sense of a blurry whiteboard sketch and still get it right.

• o4-mini is the budget-friendly genius. Built for devs who want a solid mix of speed, price, and performance. It’s cheaper than o3, but still seriously capable.

And they’re not just out here spitting text — they can actually:

• Analyze images (yep, even the messy ones from your camera roll)

• Run Python code directly in your browser (via ChatGPT’s Canvas)

• Browse the web in real time.

• Generate images and understand them in context (and can literally “think” in visuals now — wild, right?)

As of now, both models are already live for ChatGPT Pro, Plus, and Team users. Devs can also grab them through OpenAI’s APIs. 

Plus, as a bonus: There’s a special version called o4-mini-high that takes a little longer to respond but delivers even more reliable answers.

As for pricing? It’s surprisingly chill:

• o3 costs $10 per million input tokens (what you send in), and $40 per million output tokens (what it spits back)

• o4-mini on the other hand spots the same low rates as the older o3-mini which is — $1.10 for input, and $4.40 for output.

But as we all know, with great AI power comes... major freak-outs.

Because o3 and o4-mini are so much more powerful, OpenAI rolled out a brand-new "safety-focused reasoning monitor." 

Think of it as an always-on babysitter making sure the models don’t start handing out step-by-step guides for, say, making chemical weapons.

In tests, the models declined 98.7% of risky prompts — which is solid… But researchers still have their eyebrows raised at OpenAI for moving a little too fast without fully beefing up their safety nets.

(Also, weird flex but okay: One red-teaming partner said they didn’t get enough time to properly test for deception, plus, OpenAI didn’t bother releasing a full safety report for GPT-4.1, which dropped just a few days ago. So yeah, there’s a little drama happening behind the scenes.)

Zooming out:

OpenAI’s clearly trying to stay ahead of the pack.

And fun fact: o3 almost didn’t even get released inside ChatGPT. Sam Altman originally hinted they’d save it for a bigger, more advanced rollout. But with rivals moving fast, they dropped it anyway.

One Last Thing: 

OpenAI’s hinted that o3 and o4-mini might be the last solo reasoning models before the mega drop: GPT-5. So yeah… this definitely feels like the calm before another storm.

P.S.: There’s so much more under the hood, so if you're into this stuff (or building with it), I highly recommend checking out the full release details.

Thinking about starting your own newsletter? We recommend using Beehiiv — it's what we’ve used for the past 2 years and love.

If you want to support us, use our affiliate link below 👇

 🤩 Microsoft’s New AI Features!

Microsoft is on a serious mission to make AI actually useful — and not just in a cute, "write me a poem" way.

First up:

Copilot Studio just got a serious glow-up.

Yeah, Microsoft dropped a new “computer use” feature in Copilot Studio — and no, it’s not just a fancy name. This thing lets AI agents actually use your computer like a real human. Like, they can:

• Click buttons

• Select menus

• Type into fields (Basically do everything you would normally do manually — but without the coffee breaks or existential dread.)

Why does this matter?

Because now, even if there’s no API (you know, the thing apps usually need to talk to each other), Copilot can still get the job done. As the corporate vice president of Microsoft’s business & industry puts it:

“If a person can use the app, the agent can too.” (Which is both impressive and mildly terrifying...)

For now, it can actually:

• Automate boring tasks like data entry,

• Handle market research,

• Process invoices without needing messy integrations,

• Adapt to changes — that is, if a website tweaks its layout or buttons, the AI doesn’t immediately have a meltdown. It just... keeps going (Honestly, more emotionally stable than most of us — at least that’s what they claim).

Oh, and this isn’t just a Copilot Studio thing.

Microsoft also dropped a similar feature called Actions for the everyday Copilot you and I use — it can book restaurants, buy event tickets, and shop online for you.

But that’s not all:

Mustafa Suleyman, the CEO of Microsoft AI, casually announced on Bluesky that Copilot Vision — Microsoft’s AI that can literally see what’s on your screen — is now free to use inside the Edge browser.

Here’s the vibe: 

It’s a talk-based experience, meaning you just speak out loud, and Copilot will check out your screen and respond.

If you opt in, Suleyman says Copilot Vision can “literally see what you see” on your screen. He suggests using it to, for example:

• Walk you through a recipe while you cook,

• Decode confusing job descriptions,

• Even brainstorm interview prep or cover letters (although... maybe leave your actual resume-writing to your human side).

According to Microsoft’s support page, Copilot Vision can highlight parts of your screen to help you spot important info—but it won’t click links or start randomly filling out forms for you. (Which is kinda reassuring, not gonna lie.)

Quick heads up though: The full, system-wide version of Copilot Vision (outside of Edge) is still locked behind a Copilot Pro subscription.

Want to try it? Here’s how you roll:

• Open Microsoft’s website in Edge.

• Look for the prompt asking you to opt into Copilot Vision.

• Once you’re in, open the Copilot sidebar on any website, click the microphone icon, and boom — you’re live. (A little chime plays, and your browser gets a cool color shift so you know it’s working.)

Fair warning: It might be a bit glitchy.

One person testing it mentioned it took a few tries for Edge to even ask them to opt in. And once they got in? No controls showed up — just a "One moment..." message chilling awkwardly at the bottom of their screen.

(Moral of the story: Your experience might not be so smooth..)

As for privacy?

Microsoft says it logs Copilot’s responses to you, but it doesn’t collect your inputs, screen images, or page content while you're using Vision.

If you wanna stop sharing your screen, just end the session or close your browser window. Easy.

You can find the full details here.

🧱Around The AI Block

• 🎥 How to generate video scripts with AI.

• 🦾 The last 6 decades of AI — and what comes next.

• 🖼️ ChatGPT now has a section for your AI-generated images.

• 🤯 Google used AI to suspend over 39M ad accounts suspected of fraud.

• 🧠 xAI adds a ‘memory’ feature to Grok.

• 🤩 OpenAI debuts Codex CLI, an open source coding tool for terminals.

• 👩‍💻 A dev built a test to see how AI chatbots respond to controversial topics.

🛠️ Trending Tools

• Overlap turns your hour-long videos into 2-minute viral clips that capture the key moments (might still require some manual edits tho).

• AIorHuman is a interesting guessing game where you chat with a partner and decide if they're humans or AI.

• Smashing shows you curated articles with multiple perspectives on topics you care about, plus it’s from the founder of Goodreads.

• Concentric monitors who accesses your sensitive data across Microsoft, Slack and other work tools, helping prevent unauthorized sharing.

• Voice Design from ElevenLabs lets you create custom voices by typing in characteristics like “friendly Australian male” or “elderly storyteller.

🤖 ChatGPT Prompt Of The Day:

See this as an AI WoD!

Act as an academic research expert. Your task is to suggest appropriate methodologies for researching [topic]. Provide a comprehensive list of both qualitative and quantitative research methods that are best suited for the subject matter. Justify each methodology's relevance and potential advantages, ensuring they align with the research objectives. Additionally, address any potential limitations or challenges of each method, and offer potential solutions or alternative approaches. Your suggestions should be rooted in academic literature, ensuring their validity and appropriateness for academic research.

Source

Premium members get daily video prompts, premium newsletter, an no-ad experience - and more!

Already a paying subscriber? Sign In.

Premium members get::

• • 👨🏻‍🏫 A 30% discount on the AI Education Library (a $600 value - and counting!)
• • 📽️ Get the daily AI WoD (a $29.99 value!)
• • ✅ Priority help with AI Troubleshooter
• • ✅ Thursday premium newsletter
• • ✅ No ad experience
• • ✅ and more....