Introducing the first AI-native CRM

Connect your email, and you’ll instantly get a CRM with enriched customer insights and a platform that grows with your business.

With AI at the core, Attio lets you:

• Prospect and route leads with research agents

• Get real-time insights during customer calls

• Build powerful automations for your complex workflows

Join industry leaders like Granola, Taskrabbit, Flatfile and more.

👉Try Attio Pro for free

Hey there,

If you’re a developer (or just someone who loves seeing AI get smarter), you’re going to want to sit down for this first update. But even if you’re just looking for an assistant that actually knows you, Google’s latest update might finally deliver on that promise.

Let’s dive into the biggest stories, shall we?

Today:

• OpenAI & Cerebras: The Quest for Instant AI

• Gemini Finally Gets "Personal" with Your Data

• GPT-5.2 Codex: A New Heavyweight for Code

• Anthropic Snags Instagram’s Co-Founder for "Labs"

• The Massive Bet on a Universal Robot Brain

OpenAI + Cerebras: the “speed layer” for AI just got real

OpenAI announced a partnership with Cerebras to add 750MW of ultra low-latency compute to its platform, rolling online in multiple tranches through 2028.

What jumped out at me isn’t just the scale — it’s the why: OpenAI frames this as making the “request → think → respond” loop noticeably faster, especially when you’re generating code, images, or running agent-style workflows.

A bunch of outlets are also reporting the agreement is worth more than $10B.

Why this matters:

We’re past the phase where “smarter” is the only upgrade. Now it’s smarter and snappier and that changes behavior. People iterate more, ask harder things, and let agents run longer when it doesn’t feel like waiting.

GPT-5.2-Codex is now “just an API call,” and that’s a big deal for builders

OpenAI has opened GPT-5.2-Codex to developers via the Responses API, whereas it was previously limited to the Codex environment.

From OpenAI’s docs, the important practical bits are:

• It supports text + image input (text output), so you can feed screenshots/diagrams alongside instructions.

• It supports reasoning effort settings: low / medium / high / xhigh.

• Pricing listed is $1.75 / 1M input tokens and $14 / 1M output tokens (with discounted cached input).

Why this matters:

This is the difference between “cool demo tool” and “we can wire it into our pipeline.” Once it’s in the Responses API, it can sit behind code review, refactors, migration work, bug hunts — the boring stuff you never want to do twice.

Google Gemini “Personal Intelligence”: the assistant starts knowing your world

Google announced Personal Intelligence for the Gemini app — a beta (U.S. first) that connects Gemini to your Google apps with a tap (Gmail, Photos, YouTube, Search).

A few key details that are easy to miss:

• It’s opt-in, and you choose which apps get connected; you can turn it off, disconnect apps, and delete chat history.

• Rolling out to eligible Google AI Pro and AI Ultra subscribers in the U.S. over the next week; not for Workspace accounts at launch.

• Google openly mentions potential “over-personalization” mistakes (reading too much into your data) and asks users to thumbs-down bad calls.

Why this matters:

This is the real battleground for assistants: not just “knows the internet,” but knows youand has permissioned access to the messy context of your life.

🧠RESEARCH

X-Coder: Advancing Competitive Programming with Fully Synthetic Tasks, Solutions, and Tests

X-Coder is a coding model trained solely on synthetic data to master competitive programming. Using a custom pipeline to generate complex tasks and solutions, this small 7-billion parameter model outperforms larger rivals, proving that high-quality artificial data can replace real-world examples for teaching advanced reasoning.

GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughts

GlimpRouter cuts AI costs by intelligently routing queries between a cheap model and a powerful reasoning model. By letting the powerful model generate just one initial "thought" token, the system accurately gauges difficulty, sending only hard problems to the expensive model while handling easy ones cheaply.

MegaFlow: Large-Scale Distributed Orchestration System for the Agentic Era

MegaFlow is a distributed system designed to orchestrate millions of AI agents simultaneously. It addresses scalability challenges like scheduling and fault tolerance, providing a simple interface for developers to build complex, large-scale agent workflows that were previously difficult to manage efficiently.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

Can I Rank – AI-Powered SEO Software for Faster Ranking - AI‑driven SEO platform that aggregates data from multiple SEO sources and applies machine‑learning to turn it into step‑by‑step, “do this next” recommendations.

Candy AI – AI Image Editor - browser‑based image editor offering 30+ tools (38 in total) for quick tasks like converting, resizing, compressing, background removal, OCR, and more.

CandyAI – Create Your AI Girlfriend and Start Chatting - web-based platform for creating and chatting with personalized AI companions.

📲SOCIAL MEDIA

🗞️MORE NEWS

Anthropic Labs Shake-Up Instagram co-founder Mike Krieger is joining Anthropic to run their "Labs" division, a special team dedicated to building new experimental AI products. He will lead the group alongside their head of product engineering to turn their smart AI models into practical tools people can actually use. This move signals that the company wants to move beyond just making chatbots and start creating polished apps and features for the real world.

Skild AI Triples Valuation A robotics startup named Skild AI has raised a massive $1.4 billion in new cash, which triples the company's total value to $14 billion. They are building a universal "brain" that can learn to control any kind of robot, from warehouse machines to future home assistants. This huge funding from big investors like SoftBank allows them to hire more talent and speed up their goal of making robots useful in everyday life.

Kaggle Community Benchmarks Google’s data science platform, Kaggle, has launched a new feature that lets anyone create their own tests to grade how smart different AI models are. Instead of relying on official scores from big tech companies, developers can now build custom challenges to see which AI handles specific tasks best. This helps people find the perfect AI for their specific needs by looking at results from real-world experiments rather than just standard exams.

Changes at Thinking Machines Lab Mira Murati, the former technology chief at OpenAI, has launched a new company called Thinking Machines Lab that is already valued at $12 billion. Rather than just building bigger and more expensive AI brains, her team is releasing tools that help developers tweak existing models to be smarter and more efficient for specific jobs. Their first product, called Tinker, makes it easy to customize open AI models without needing a supercomputer to do it.

ChatGPT Translate Tool OpenAI has quietly released a dedicated translation page that competes directly with Google Translate, allowing users to swap text between languages with high accuracy. Unlike older translation tools that can sound robotic, this new feature uses advanced AI to understand the subtle context and culture behind the words. It is currently free and aims to make communication across different languages feel much more natural and human.

Microsoft and Anthropic Partnership Despite being the biggest backer of OpenAI, Microsoft has signed a major deal to sell AI models from rival company Anthropic to its own business customers. In exchange, Anthropic has agreed to spend billions of dollars renting Microsoft’s cloud servers to power their technology. This partnership shows that Microsoft wants to offer every top-tier AI option to its clients, even if it means working with its partner's biggest competitor.