OpenAI Makes Premium AI 80% Cheaper with o3-Pro

Hello Engineering Leaders and AI Enthusiasts!

This newsletter brings you the latest AI updates in a crisp manner! Dive in for a quick recap of everything important that happened around AI in the past two weeks.

And a huge shoutout to our amazing readers. We appreciate you😊

In today’s edition:

🔧 OpenAI rolls out o3-pro with major price drop
🌐 The Browser Company bets big on AI with Dia
🤖 Meta builds AI that “thinks before acting”
💻 Gemini 2.5 Pro gets secure coding upgrade
🧬 Sakana’s AI learns to rewrite itself
🧠 Knowledge Nugget: Thoughts on How AI Will Shape Cyber Security by Utku Sen

Let’s go!

OpenAI rolls out o3-pro with major price drop

OpenAI released o3-pro, an advanced reasoning model now available to ChatGPT Pro and Team users. It outperforms previous o3 versions and rivals’ top models from Google and Anthropic, while slashing costs by 80%.

o3-pro is built to handle deeper reasoning and long-context tasks like math, coding, and scientific analysis. Evaluators consistently rated it higher than o3 across domains. It also supports tool use (web, files, code interpreter) but doesn’t yet include image generation or Canvas support.

Why does it matter?

With o3-pro, OpenAI is optimizing usability and cost. By making deep reasoning tools cheaper and faster, this move pressures rivals to rethink how they package frontier capabilities for real-world teams.

Source

The Browser Company Bets Big on AI With Dia

The Browser Company dropped Dia, an AI-first browser that embeds models into the browsing experience. Users can query, summarize, auto-code, or interact with content using agents that overlay directly onto web pages. It’s designed for productivity and to streamline everyday workflows.

Instead of bookmarks or tabs, Dia organizes sessions by goals. Agents can co-browse, trigger actions, and respond inline, similar to what extensions try to do, but natively baked in.

Why does it matter?

Browsers are becoming battlegrounds for AI integration, and Dia is the clearest signal yet that the interface layer is up for grabs. As AI agents shift from chat boxes to embedded workflows, Dia shows what it looks like when the web itself becomes programmable. Should we expect a wave of competition from incumbents chasing the same vision?

Source

Meta builds AI that “thinks before acting”

Meta has released V‑JEPA 2, a vision-based world model trained on video to simulate physical environments. It helps AI predict the outcomes of actions, like where objects will land, without trial and error. It’s already in use in Meta’s robotics labs, powering arms that can pick, place, and plan with foresight.

V-JEPA 2 runs 30x faster than Nvidia’s Cosmos (by Meta’s benchmarks) and aims to reduce the need for huge robotic datasets. Meta sees it as a step toward real-world AI assistants that can reason and act with human-like foresight with no labels or trial-and-error training required.

Why does it matter?

From AI that reacts to AI that mentally simulates the world before acting—this is a core ingredient of human-like intelligence. Meta may be laying the groundwork for more adaptable AI systems that can act with foresight, not just react to prompts.

Source

Gemini 2.5 Pro Gets Secure Coding Upgrade

Google just released a preview update to Gemini 2.5 Pro, this time focusing on safe and compliant AI coding. The model includes sandboxing, static analysis, and risk-aware guardrails, targeting enterprise use in regulated environments.

The update supports traceable, auditable code generation and addresses concerns from enterprise dev teams around reliability and safety. It’s designed for workflows in healthcare, finance, and other high-stakes domains, and rolls out through AI Studio, Vertex AI, and the Gemini app.

Why does it matter?

Just weeks after rolling out its coding-focused 2.5 Pro I/O update, the tech giant is back with another upgrade featuring widespread quality improvements. Google continues to lead the benchmarks and is shifting its release strategy now, favoring frequent ‘preview’ releases ahead of full model launches.

Source

Sakana’s AI Learns to Rewrite Itself

Sakana AI and the University of British Columbia have introduced the Darwin Gödel Machine (DGM), an AI agent that autonomously upgrades its own code to perform more effectively. It improves task results by up to 150% without requiring human intervention. Starting as a coding assistant, DGM added its tools like error memory and peer review, boosting SWE-bench from 20% to 50%, and Polyglot from 14% to 30%.

The system is inspired by biological evolution. DGM mutates parts of its code, tests outcomes, and archives what improves performance. Unlike fine-tuned LLMs, DGM’s upgrades worked across different models, suggesting its gains weren’t tied to any one foundation model.

Why does it matter?

DGM moves beyond fine-tuning toward self-evolving AI. This opens the door to faster, more adaptive systems, but also raises urgent questions about how we monitor and govern self-modifying agents.

Source

Enjoying the latest AI updates?

Refer your pals to subscribe to our newsletter and get exclusive access to 400+ game-changing AI tools.

Refer a friend

When you use the referral link above or the “Share” button on any post, you’ll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.

Knowledge Nugget: Thoughts on How AI Will Shape Cyber Security.

In this article, Utku Sen argues that the security industry is on the brink of a structural shift driven not by threats, but by tools. Traditional roles like penetration tester or cloud security analyst are being chipped away by LLMs that can analyze code, flag risks, triage vulnerabilities, and automate response actions. Even interfaces like Splunk or CrowdStrike can now be queried via agents, turning complex dashboards into plain English instructions.

The implications are broad. DevSecOps roles could shrink into generalist “AI overseers.” Vulnerability scanners may become background services managed by workflows, not humans. And as LLMs outperform legacy security tools, demand may shift from specialists to strategic conductors who know how to direct AI systems, not run them line by line.

Why does it matter?

Security has long been reactive and human-dependent. But with AI taking over judgment-based tasks, the center of gravity is moving. Engineers will need less domain depth and more orchestration skills. Meanwhile, attackers are getting the same upgrades, meaning patch windows could shrink from weeks to minutes.

Source

What Else Is Happening❗

🧠 Meta forms new Superalignment team to build safe superintelligence, tapping top researchers and focusing on long-term AI control.

🏛️ The UK government will use Google’s Gemini to speed up housing and infrastructure planning decisions, aiming for improved efficiency.

📁 ChatGPT now integrates with Google Drive and Dropbox, offering automated meeting notes with real-time file summarization.

🛡️ Anthropic launches Claude Gov, a secure AI system tailored for U.S. national security agencies with classified data access controls.

🎥 Bing launches AI-powered Video Creator for text-to-video generation, now in preview with 100+ scene templates and voiceovers.

🧬 Clairity, an AI tool for detecting breast cancer in mammograms, gets FDA clearance, boosting early detection and diagnostic accuracy.

🎨 Heygen debuts AI Studio, letting creators generate branded video content with avatars, voice cloning, and scene-based editing tools.

📊 Meta will use AI to fully automate ad creation, from copywriting to visual design, optimizing ads in real time based on performance.

🧪 New study finds AI can predict a person’s age from a single blood sample with surprising accuracy, raising fresh bioethics questions.

New to the newsletter?

The AI Edge keeps engineering leaders & AI enthusiasts like you on the cutting edge of AI. From machine learning to ChatGPT to generative AI and large language models, we break down the latest AI developments and how you can apply them in your work.

Thanks for reading, and see you next week! 😊

Read More in  The AI Edge