China’s AI Breakout: The Week DeepSeek Surpassed the West

Welcome back apprentices! 👋

Let’s be honest — most of us gave up on math after high school. 

But what if an AI just smoked the best human brains at the world’s hardest math tests… and then offered you its notes for free? 

Something big just happened in AI land.

And the usual suspects (OpenAI, Google) weren’t invited.

In today's email

• AI just crushed elite math tests

• It’s 10x cheaper than GPT-5

• China now leads in AI downloads

• Time to rethink your AI tools?

• Monetization Tool

_{Read Time: 4 minutes}

Quick News 

🧠  Mathletes, meet your new teammate: AI. Harmonic’s AI system, Aristotle, just cracked a 30-year-old Erdős problem — in six hours — then checked its own proof in under a minute using Lean. This isn’t just smart; it’s a new era researchers are calling “vibe proving” — think AI intuition meets formal logic, with receipts. Backed by $120M and flexing with an IMO gold, Aristotle's not just doing the homework — it's rewriting the whole syllabus.

🚨 Code Red Button. The tables have turned in the AI arms race. After Google’s Gemini 3 upgrade, OpenAI CEO Sam Altman has declared a “code red” sprint to boost ChatGPT, delaying ads and agents to double down on user experience. A faster reasoning model, Shallotpeat, drops next week (and reportedly beats Gemini), while a major update codenamed Garlic is cooking for 2026 — think GPT-5.2 on steroids. Three years ago, Google panicked over ChatGPT. Now, it’s OpenAI playing catch-up.

🧬 AI Cracks the Genetic Code. Harvard Medical School’s new tool, popEVE, is scanning entire genomes and solving medical mysteries that stumped doctors for years. By comparing human DNA with hundreds of thousands of species, it’s flagging fewer false alarms — dropping scary variant alerts from 44% to just 11%. In one study, it cracked a third of previously undiagnosed developmental cases and even spotlighted over 120 genes no one had on the radar.

Together with Guidde

Effortless Tutorial Video Creation with Guidde

Transform your team’s static training materials into dynamic, engaging video guides with Guidde.

Here’s what you’ll love about Guidde:

1️⃣ Easy to Create: Turn PDFs or manuals into stunning video tutorials with a single click.
2️⃣ Easy to Update: Update video content in seconds to keep your training materials relevant.
3️⃣ Easy to Localize: Generate multilingual guides to ensure accessibility for global teams.

Empower your teammates with interactive learning.

And the best part? The browser extension is 100% free.

Check out Guidde

DeepSeek 
How DeepSeek Is Schooling the AI Giants

The Chinese AI lab that’s been dropping banger models all year, just released Math-V2, a new open-source MoE model that nailed 5 out of 6 problems from the 2025 International Math Olympiad and nearly aced the Putnam — the toughest college math exam out there. That’s not just “AI doing math,” that’s AI outdoing humans.

But wait, there’s more: DeepSeek’s newest V3.2 and V3.2-Speciale reasoning models now rival GPT-5, Gemini 3 Pro, and Claude 4.5 in key benchmarks, at a tenth of the cost. And a new MIT–Hugging Face report shows that Chinese labs — led by DeepSeek and Alibaba’s Qwen — have overtaken the U.S. in open AI downloads. 

Translation? The AI center of gravity is tilting east — and fast.

Key Points That’ll Make You Sound Smart at Lunch

• 🥇 DeepSeek-Math-V2: Scored 118/120 on the Putnam exam — higher than the best human score.
  

• 🔍 IMO 2025 Performance: Solved 5 out of 6 problems — effectively gold-medal territory.
  

• 🧠 IMO ProofBench: Scored 61.9%, nearly matching Google’s Gemini Deep Think, and tripling GPT-5’s 20%score.
  

• 🔄 Verifier-Generator Loop: Uses a built-in proof-checker that assigns confidence to each logical step — like pair programming with a philosopher.
  

• 📉 Cost Drop: V3.2 offers $0.28 input / $0.42 output per 1M tokens — compare that to GPT-5.1’s $1.25 / $10, or Claude’s $3 / $15.
  

• 🧾 Licensing: MIT-licensed and hosted on Hugging Face, unlike most Western "open" models that feel more like freemium trials with API keys.
  

• 🌐 Download Power Shift: Chinese labs now hold 17.1% of Hugging Face downloads vs. the U.S.’s 15.8% — a full reversal since 2022.
  

• 📉 Transparency Crisis: Only 39% of open models disclose their training data today — down from 79.3% in 2022.

The MoE Advantage – Smarter, Cheaper, Actually Open

Let’s talk architecture. 

Mixture-of-Experts is like a dinner party: instead of inviting every expert to every question, you only bring the ones who can actually help. 

That’s what DeepSeek is doing. Only parts of the model are activated per task, reducing computation and latency without sacrificing quality.

Why this matters:

• MoE lets DeepSeek cram 685 billion parameters into a usable, efficient, and cheap system.

• And they’re not hoarding it — it’s open-source under MIT license, no strings or signup walls.

• That’s a massive shift from OpenAI’s “You can look but not touch” approach to LLMs.

If you’re building internal tools, research workflows, or math-heavy apps, you no longer need a seven-figure API budget. Clone Math-V2, fine-tune it on your dataset, and you’ve got a model that explains proofs better than most grad students — at a fraction of the price.

Geopolitics of AI – The New Digital Silk Road

Forget TikTok bans — the real shift is happening in AI downloads. 

According to the new MIT x Hugging Face study:

• DeepSeek + Alibaba’s Qwen captured 14.2% of global model downloads from August 2024 to 2025.

• U.S. industry? Down to 15.8%, with ComfyUI, not OpenAI or Meta, leading the charge with just 5.4%.

• Google, Meta, and OpenAI — which owned 40%+ of Hugging Face downloads before 2023 — are gone from the leaderboard.

The punchline? 

The U.S. once led open AI by a mile.

Now it's China bringing the best models to your dev folder.

Price Pressure on U.S. APIs – Time to Rethink Your Stack

Here’s a quick math check:

That’s not just a difference. That’s business-model-breaking.

DeepSeek isn’t just matching performance — it’s flipping the economics. 

And when those models are MIT-licensed and locally deployable? 

That changes who builds, who scales, and who wins.

Takeaway: Audit your LLM bill. 

If your app’s cost-per-query is creeping into the red, it might be time to swap the big-name APIs for something leaner, meaner, and just as smart.

What’s the Deal for You?

You no longer need to settle for API rate limits, tiered access, or proprietary tools that gatekeep their best features. 

The open-source AI wave isn’t just surviving — it’s surpassing the best of Silicon Valley. And for the first time in a long time, innovation might actually be free.

Help Your Friends Level Up! 🔥

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and you’ll win a little surprise 👀

Build Your AI Income

Turn AI into Your Income Engine

Ready to transform artificial intelligence from a buzzword into your personal revenue generator

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

• A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

• Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

• Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

🧪 Test the Prompt

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

👑 The winner is…

Last week, we challenged you to test GPT-4o’s visual generation skills with this prompt.

Here’s the WINNER:

Congrats again to David for his creation!🥳

Want to be featured next? Keep those generations coming!

🎨 Prompt: The Datafall Observatory

Deep in a mountain carved entirely from mirrored stone and fiber-optic veins lies a vast cylindrical chamber known as the Datafall Observatory. From the ceiling descends a colossal waterfall of pure information—millions of luminous strands cascading like liquid glass, each thread carrying shifting symbols, blueprints, and glowing particles. At the center of the chamber is a raised platform where the falling data converges into a single, ultra-detailed, photorealistic construct of [your creation], forming itself in real time as if the universe is rendering your idea from raw code. The camera frames the scene from a dramatic mid-distance angle, capturing razor-sharp reflections, high-contrast lighting, and the mesmerizing clarity of the data-stream illuminating the dark, vaulted space.

We’ll be featuring the best generations in our next edition!

Want to start your own Newsletter?

Make Newsletter Magic in Just Minutes

Your readers want great content. You want growth and revenue. beehiiv gives you both. With stunning posts, a website that actually converts, and every monetization tool already baked in, beehiiv is the all-in-one platform for builders. Get started for free, no credit card required.

_FEEDBACK