Welcome to this week's edition of Overclocked!

From OpenAI’s state-of-the-art reasoning and semi-agentic models to Microsoft’s revival of Recall, the last week of AI news delivered a flood of information. We’ll also show you some of the coolest free AI tools in Meta Labs and test your ability to use the right AI tools at the right time. Let’s break it all down ⬇️

In today’s newsletter ↓
🤖 OpenAI gives access to o3 & o4‑mini to most users
⚡ Meta labs makes using AI fun and easy
🎞️ EU considers even stricter data laws
🎥 Adobe Firefly to allow unlimited pictures and gifs
🔥 AI Tool Match Quiz

🌪️ A Whirlwind of Powerful AI Releases And Revivals

🫨 o3 & o4 mini are seriously breaking the internet

OpenAI’s twin drop stole early‑weak headlines: o3—its most advanced reasoning model yet, excelling at complex coding and structured reasoning tasks—and the streamlined o4‑mini, optimized for faster responses and lower computational demands. 

Both models now enjoy full ChatGPT integration with expanded tools like web browsing, Python scripting, and file analysis, significantly boosting their practical versatility. Many users are already comparing o3’s reasoning capabilities to those of Claude and Gemini, citing its performance in data-heavy tasks and chain-of-thought reasoning as standout features.

🎳 Google strikes back in less than one day

Just 24 hours later, Google unveiled Gemini 2.5 Flash, emphasizing low latency and cost efficiency. Developers can toggle an internal chain-of-thought capability, allowing Gemini Flash to handle complex prompts requiring deeper reasoning and nuanced understanding without significant performance drawbacks. This model is optimized for quick responses and minimal resource usage, making it ideal for scalable deployment in enterprise settings.

😎 Recall returns (again)

Microsoft’s controversial Windows Recall feature re‑emerged as an opt‑in preview for Windows Insiders. After nearly a year of privacy revisions, Recall now aims to provide users with enhanced productivity while upholding data integrity. The feature logs user activity in a searchable format, allowing users to revisit previously viewed content—but with new guardrails in place after regulatory backlash.

Learn more about Windows Recall

👊 Meta vs. Apple

Adding drama to the week, Meta blocked Apple’s on‑device “Apple Intelligence” features inside Facebook, Instagram, and Threads on iOS, heightening speculation ahead of WWDC 25.

The decision hints at deeper competitive tensions, especially as both companies prepare to make AI central to their mobile ecosystems. Developers are watching closely to see how this will affect cross-platform functionality and third-party app capabilities.

🧪 Let’s Not Forget About These Free Epic AI Experiments From Meta

Speaking of Meta, it recently refreshed its public AI Demos Hub—and it’s packed with toys you can use right in the browser. It’s been out for a while, but not many people know about it. We took the four coolest for a spin. Here’s what we learned:

🖼️ Segment Anything 2 might be one of the coolest features ever

Think Photoshop’s magic wand on steroids. SAM‑2 refines Meta’s segmentation model with a bulk-mask predictor that outlines dozens of objects in a single click. It rarely misses an edge. Perfect for removing clutter or isolating details.

 → Try Segment Anything 2

🌐 Hear what you sound like in another language

Building on SeamlessM4T and Seamless Expressive, this demo turns live speech into another language—and keeps your voice and tone. It supports 100+ languages, works in noisy settings, and translates with under two seconds of latency.

→ Test Seamless Translation

✏️ Give your animated drawings life

Upload any sketch and Meta will rig it with a digital skeleton, then animate it in seconds. Choose from pre-set loops, keyframes, or even dance moves. Teachers and content creators are already using it for quick visuals.

→ Animate Your Sketch

🔊 Take audio generation to another level

Meta’s unified text-to-audio model generates Foley, ambient sound, and character voices. You can clone your own voice with a 3-second clip or create environments like "noir cafe" in one prompt.

→ Play with AudioBox

The Weekly Scoop 🍦

🖥️ Nvidia to produce $500B in US AI servers with partners

🦙 Meta drops Llama 4 Scout & Maverick multimodal models

🏥 xAI drops Grok Studio for free

🚀 Amazon Bedrock adds Pixtral and Nova Sonic to lineup

🛡️ EU eyes stricter cyber rules ahead of AI Act rollout

🎨 Adobe's Firefly AI now offers unlimited gen content

❓ AI Tool Match Quiz

🛠️ Tools in play
A) Perplexity Pro – research copilot that cites sources on demand
B) Gemini 2.5 Pro – Google’s multimodal reasoning powerhouse
C) OpenAI o3 – best‑in‑class chain‑of‑thought coder & math whiz
D) Claude Sonnet – long‑context writing maestro with friendly tone
E) Pika 1.0 – lightning‑fast text‑to‑video generator
F) Midjourney v7 – stylistic text‑to‑image sorcerer
G) ChatGPT 4o – multimodal wizard with impeccable image generation

Scenarios:

1. On‑Demand Source‑Backed Research
   You need a concise, 200‑word briefing on the latest clinical trial results for Alzheimer’s treatments, complete with footnoted citations—all generated in under 20 seconds. Which tool do you choose?

2. Multimodal Puzzle Solver
   You’ve got a PDF of architectural blueprints plus aerial photos, and you need a step‑by‑step plan highlighting structural risks and suggesting reinforcement strategies. Which tool handles both text and images with deep reasoning?

3. Chain‑of‑Thought Coding & Math
   A complex algorithm is failing silently in your Python script, and you need inline annotations, intermediate variable dumps, and a clear, copy‑and‑pasteable fix. Which tool’s coder‑centric reasoning do you call?

4. Extended Long‑Form with Human Touch
   You’re commissioning a 10,000‑word whitepaper on sustainable supply chains, ensuring a friendly yet authoritative tone and seamless transitions across four chapters. Which model’s long‑context prowess is best?

5. Lightning‑Fast Text‑to‑Video
   You want a 30‑second explainer video—animated product demo with voiceover—ready in under a minute straight from your script. Which tool do you spin up?

6. High‑Fidelity Stylized Imagery
   You need a series of surreal, painterly concept art pieces for your fantasy novel cover, each in a distinct style. Which text‑to‑image sorcerer is your go‑to?

That's all for this week's Overclocked! What latest release surprised you the most, and what cool things will you create with Meta labs? Reply and let us know—we love hearing your thoughts!

Zoe from Overclocked