🐋 DeepSeek unveils new AI models

Good morning, AI enthusiast. China’s DeepSeek is back in the global AI race — launching two new models that claim GPT-5-level performance and Gemini-3.0-Pro-level reasoning.

In today’s AI newsletter:

• DeepSeek launches V3.2 & V3.2 speciale

• OpenAI declares code red

• Apple names new AI Chief

• How to turn any gitHub repo into interactive code dashboard

• AI tools & more

AI MODELS

🤖 DeepSeek Launches V3.2 & V3.2 Speciale

DeepSeek unveiled two new AI models, DeepSeek-V3.2 and DeepSeek-V3.2 Speciale, designed to deliver high-efficiency reasoning, advanced tool-use, and agent-friendly workflows. After months of silence, the company is signaling it’s ready to compete directly with OpenAI and Google again.

• DeepSeek-V3.2 integrates reasoning directly into tool-use, working in both “thinking” and “non-thinking” modes

• Speciale edition pushes max reasoning, rivaling Gemini-3.0-Pro

• Can generate huge agent-training datasets across 1,800+ environments and 85,000+ complex tasks

• Uses DeepSeek Sparse Attention (DSA) for faster long-context performance

• V3.2 reportedly matches GPT-5 in balanced inference; Speciale hits Olympiad gold-medal reasoning benchmarks

If these results hold up, DeepSeek V3.2 signals a major shift, open-source models are no longer just "toy" alternatives. Better training recipes, stability engineering, and RL can close gaps without 10× the compute, increasing competitive pressure on closed-weight labs and accelerating innovation across the ecosystem.

AI NEWS

🚨 OpenAI Declares Code Red

OpenAI is going all-in on upgrading ChatGPT, with a new reasoning model set to launch next week that reportedly outperforms Google’s Gemini 3 in internal tests. The company is reallocating people, compute, and priorities to sharpen the product before the next wave of competition hits.

• “Code red” shifts resources away from ads, AI agents, Pulse, and other experiments

• New reasoning model (powering Thinking Mode + Deep Research) scores above Gemini 3 in benchmarks

• Ads and monetization rollout delayed to focus on UX, personalization, and reliability

• ChatGPT upgrades target faster responses, fewer unnecessary refusals, and better model behavior

• Image generation gets top priority after Google’s Nano Banana Pro raised the bar

With Google and Anthropic closing in, OpenAI is refocusing on the core product used by 800M+ people every week. If the new reasoning model truly beats Gemini 3, ChatGPT could regain an edge in both public benchmarks and everyday usage, just as the AI platform wars enter their most competitive phase yet.

AI NEWS

🍏 Apple Names New AI Chief After

Apple has replaced John Giannandrea, its AI chief since 2018, with Amar Subramanya, a senior Microsoft leader who previously spent 16 years at Google working on Gemini Assistant. The move comes after a year of embarrassing misfires with Apple Intelligence and the delayed Siri overhaul.

• Giannandrea is stepping down, staying only as an adviser until spring

• Amar Subramanya (ex-Google, ex-Microsoft) now leads Apple’s AI org

• His hiring signals Apple wants deep Google/Gemini expertise to catch up fast

• Apple Intelligence stumbled with false notifications and unreliable features

• Siri’s big AI upgrade was so broken it was delayed indefinitely

This is one of Apple’s biggest AI resets yet. Competitors like OpenAI, Google, and Meta are shipping massive cloud-based models, while Apple’s privacy-first, on-device approach has left it trailing. Subramanya now faces a massive challenge: fix the culture, stop the talent drain, and rebuild Siri into a competitive AI assistant, all while Apple’s rivals sprint ahead.

HOW TO AI

🧠 How to Turn any GitHub Repo into Interactive Code Dashboard

In this tutorial, you’ll learn how to transform any open-source GitHub repository into a complete visual knowledge hub using Google’s new CodeWiki platform, a tool that automatically generates diagrams, explanations, relationships, and even short videos so you can understand any codebase in minutes.

🧰 Who is This For

• Developers onboarding to unfamiliar codebases

• Open-source contributors studying large projects

• Founders exploring technical architecture before adoption

• Anyone who wants instant clarity from complex GitHub repositories

STEP 1: Access the CodeWiki Platform

CodeWiki is available as a public preview. It’s built to help you explore and understand open-source projects instantly.

Go to official site at CodeWiki.

The interface is extremely simple, just a single page waiting for your repo link.
Since this preview only supports public projects, make sure the repository you want to explore is openly accessible.

STEP 2: Enter Your GitHub Repository URL

First copy the full URL of the GitHub repo you want CodeWiki to analyze.
They come to codewiki and paste it into the input box in middle of the homepage and hit enter.

Once you submit, Gemini starts scanning the entire codebase. It reads the files, functions, classes, and dependencies, then converts everything into a structured knowledge graph.

STEP 3: Explore

Within a few seconds, you’ll see a complete interactive breakdown of the code.

A guided overview explains the project’s purpose and architecture.

Move into the diagrams section to see automatically generated flowcharts, class diagrams, and dependency maps.

These visuals show how the code pieces fit together, making it easy to understand even massive repos.

For complex sections, CodeWiki also creates short video explainers that walk through how certain parts work. Each explanation links directly to the exact source code line it’s referencing.

STEP 4: Chat Directly With Your Codebase

On the right side of the screen, you’ll see a chat panel powered by Gemini. This is where CodeWiki becomes magic.

Ask natural-language questions like “How does this repo handle authentication?” or “Where is the data processing pipeline defined?”

The chatbot replies instantly with precise, code-aware answers. It also links you to the specific file or line inside the generated wiki.

This lets you understand unfamiliar projects without manually digging through hundreds of files.

Nvidia announces Alpamayo-R1, an AI model for autonomous driving research, and calls it the “first industry-scale open reasoning vision language action model”.

Samsung unveils the Galaxy Z TriFold, with a 6.5" outer screen, a 10" inner screen, a Snapdragon 8 Elite for Galaxy SoC, and a 3.9mm body at its thinnest point.

Elon Musk’s xAI is reportedly set to raise a $15B round of funding at a $230B pre-money valuation next month, CNBC reported.

Marques Brownlee plans to shut down his wallpaper app Panels on December 31; Appfigures: it had 900K downloads and $95K in user spending since its 2024 launch.

🍌 Nano Banana 2: Google’s Gemini 3 image model with perfect text and character consistency

📸 Perplexity: Now lets you virtually try on products while you search

🧮 Math V2 by DeepSeek: A smarter, open-source model for tough math

🤖 Claude Opus 4.5: Anthropic’s new flagship model for coding, agents, and real computer use

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas