In partnership with
👋 Hey builders,
Something shifted this week. China released two open-source AI models that run neck-and-neck with America's best. Moonshot AI dropped Kimi K3 with 2.8 trillion parameters. Three days later, Alibaba answered with Qwen 3.8 at 2.4 trillion. Both are going open-weight. Both cost a fraction of what GPT and Claude charge. And both showed up while US lawmakers are still arguing about whether to regulate or restrict.
The New York Times ran a piece on Sunday titled "Why Silicon Valley Can't Stop Looking Over Its Shoulder at China." The BBC covered it. Forbes covered it. This is not a niche story. This is the story.
And it gets personal. Google's AI Mode is now swallowing 60% of searches without sending a single click to the websites that created the content. Publishers are hemorrhaging traffic. This newsletter? It depends on the open web too. So today we learn how to navigate a world where open-source AI models are suddenly your best option, and the open web that hosts them is under siege.
Let's get into it.
HERE'S WHAT HAPPENED IN AI TODAY
Moonshot AI released Kimi K3, the largest open-source AI model ever built
Alibaba previewed Qwen 3.8, claiming it trails only Anthropic's Fable 5
Google's AI Mode now handles 60% of searches without a click to publishers
Today's skill: how to pick the right AI model for any task
LEARN AI
China Just Caught Up
For two years, the story was the same. American labs build the frontier. Chinese labs copy it. The gap slowly shrinks but never closes. That story ended this week.
Moonshot AI, a Beijing startup backed by Alibaba, launched Kimi K3 on July 16. It has 2.8 trillion parameters. It uses a Mixture-of-Experts architecture with 896 experts, activating just 16 per token. It has a 1-million-token context window. And on the Artificial Analysis Intelligence Index, it scored 57.1, against GPT-5.6 Sol at 58.9 and Fable 5 at 59.9. In web interface engineering, it ranked first, beating Fable 5 in blind human-preference tests. The full open-source weights drop on July 27.
The demand was so intense that Moonshot had to pause new subscriptions. The GPUs could not keep up.
Then, three days later at the World Artificial Intelligence Conference in Shanghai, Alibaba previewed Qwen 3.8. Two point four trillion parameters. Multimodal: text, images, video, documents. Alibaba claims it is "second only to Fable 5." No independent benchmarks yet. No model card. No Hugging Face listing. The open weights are promised "soon" with no date. The claim without the receipts is unusual for Alibaba, whose previous flagships shipped with full benchmark results.
But here is what matters more than the benchmarks. These models are open-source and cheap. Kimi K3 costs $3 per million input tokens and $15 per million output tokens, with cache-hit input at just $0.30. Fable 5 and GPT-5.6 cost many times that. Pinterest, Airbnb, and other Fortune 500 companies are already using Chinese models because they are, as Airbnb CEO Brian Chesky put it, "very good, fast, and cheap." A BBC analysis found that open-source techniques using Chinese models are 30% more accurate than leading off-the-shelf American models for certain tasks, at 90% lower cost.
The NYT's Cade Metz captured the dynamic on Sunday: "Good enough often beats best if it's a lot cheaper. That has been true since PCs replaced mainframes, and it could happen again with AI."
Xi Jinping hailed Beijing as the champion of a "new global AI order" in a speech on Friday. The framing is intentional. China is not just catching up. It is redefining the rules. Open-source is the weapon. Affordability is the strategy. Adoption is the victory.
Why this matters: If you are choosing an AI model this month, the calculus just changed. You no longer need to pay premium prices for frontier performance. Kimi K3 proves that open-source models can match closed ones on real tasks. Qwen 3.8, once its weights arrive and benchmarks are verified, could do the same. For developers, startups, and companies watching their AI budgets, this is the moment to reevaluate. For the US labs spending hundreds of billions to maintain a lead that is now measured in single-digit percentage points, this is a warning.
Our take: The open-source AI model wave from China is not a threat to American innovation. It is a reality check on American pricing. When a free model scores within 5% of a paid one on independent benchmarks, the paid model needs to justify its cost with something other than raw intelligence. Speed, reliability, privacy, support, and ecosystem integration are where the real competition moves next. The models are converging. The experience around them is what will separate the winners. And for learners, this is the best time ever to experiment. Download Kimi K3 on July 27. Try Qwen 3.8 when the weights drop. Run the same prompt across GPT, Claude, and the open alternatives. You will learn more about AI in one afternoon than most people learn in a month of reading headlines.
From Our Partners
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning
What is an EOR—and why are companies using it?
Opening entities in every country can be slow, expensive, and hard to scale.
That's why more companies are using EOR to hire globally faster.
See how Oyster helps teams hire, pay, and support talent in 180+ countries while staying compliant along the way.
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.
AI SKILL OF THE DAY
How to Pick the Right AI Model for Any Task
Ten different AI models are now available for free or cheap. Most people pick one and stick with it. That is like using a hammer for everything because you own a hammer. Here is a simple framework for choosing the right tool for the right job.
The 4-question model picker:
What is the task? Writing, coding, research, data analysis, image generation, or conversation? Different models excel at different things. Claude for long-form writing. GPT for broad ecosystem and plugins. Gemini for multimodal. Kimi K3 for long-context coding. Qwen for enterprise tasks with multimodal needs.
What is your budget? Free tier gets you far. Kimi K3 chat is free. Qwen preview is free. ChatGPT free tier is powerful. If you are paying more than $20 a month, make sure you are using features that justify it.
Do you need privacy? If you are handling sensitive data, closed models send your data to servers. Open-source models like Kimi K3 can be self-hosted. You control the data. That matters for healthcare, legal, and finance work.
How fast do you need it? Smaller models respond faster. If you need quick turnarounds on simple tasks, a lightweight model beats a frontier one. Use the big guns for complex reasoning, not for reformatting a list.
Copy the prompt below, paste it into any AI chat tool, and fill in your real needs.
Help me pick the best AI model for my needs.
My main tasks: [LIST 3-5 THINGS YOU USE AI FOR]
My budget: [FREE / UP TO $20/MONTH / NO LIMIT]
Privacy needs: [NONE / MODERATE / MUST SELF-HOST]
Speed preference: [FAST RESPONSE / DEEP THINKING / BOTH DEPENDING ON TASK]
Give me:
1. The best model for each of my tasks and why
2. Whether I can use a free option instead of paying
3. One task where I should switch from my current model
4. One task where I should NOT switch
Keep it practical. No jargon. I just want to know which tool to open when.The key insight: There is no single best AI model anymore. There is a best model for each task. The people who get the most from AI are the ones who match the tool to the job, not the ones who pay the most for one subscription.
TREATS TO TRY
TREATS TO TRY
Kimi K3 (Moonshot AI) - The 2.8T open-source model that just matched frontier benchmarks. Try it free at kimi.moonshot.cn. Full open weights drop July 27. Free to download, run, and customize. Try it
Qwen 3.8 (Alibaba) - The 2.4T multimodal model that claims second place behind Fable 5. Live preview now at chat.qwen.ai. Open weights promised soon. No benchmarks yet, so take the claims with a grain of salt until the receipts arrive. Check it out
Kagi Search - Ad-free search engine with AI overviews turned off by default. If Google's AI Mode is driving you nuts, this is the antidote. $10/month. Learn more
AROUND THE HORN
Google's AI Mode is killing the open web. Sixty percent of searches now end without a click. Publisher traffic is down 33% globally. HubSpot lost 70-80% of its organic traffic. Google says AI Overviews generate more clicks. Independent data says otherwise. This affects everyone who creates content online.
Xi Jinping hailed Beijing as champion of a "new global AI order" in a speech Friday. The message: China is not just participating in the AI race, it is rewriting the rules around open-source access and affordability.
Moonshot AI paused new Kimi K3 subscriptions because demand overwhelmed GPU capacity. When you have to turn away users because your model is too popular, that is a signal.
Meta's Oversight Board found all major LLMs refuse to criticize restrictive governments. Tested across Anthropic, DeepSeek, Google, Meta, and OpenAI models. AI alignment with political power is a pattern, not a bug.
Alibaba's Qwen 3.7 Max was API-only. Qwen 3.8 returning to open weights reverses course. As one AI researcher noted, "Something is changing." The open-source arms race among Chinese labs is accelerating.
FREE RESOURCES
1000 Claude AI Prompts That Feel Illegal To Know
Most people do not get bad results from Claude because the AI is weak. They get average results because their prompts are unclear. They rewrite. They try random phrasing. They waste time fixing weak outputs. And they keep starting from scratch.
That is the problem this library solves.
1,000 numbered Claude prompts across 20 chapters
40 master prompt templates and 6 prompt frameworks
Covers writing, marketing, business, coding, research, productivity, and automation
Free starter pack available. Premium edition for the full library
1,076 copies sold. Rated 4.9/5
ARTICLE OF THE DAY
10 AI Roles Companies Are Hiring For Right Now
The highest-paying AI jobs are not technical. They bridge AI and business. Strategy, operations, and product roles pay as well as engineering.
AI job postings are 3x what they were in 2024. The demand is real and growing, especially for roles that combine domain expertise with AI fluency.
The 4-step path: Pick a lane, build in public, create proof, speak the language. You do not need a computer science degree. You need to show you can use AI to solve real problems.
Open-source AI models are creating new roles. Model evaluator, AI implementation specialist, and open-source AI engineer are titles that barely existed a year ago.
The people getting hired are the ones who can prove they can use the tools, not just talk about them.
Thank you for reading. The AI race used to be about who could build the smartest model. That race is now a tie measured in single digits. The next race is about who can make AI the most useful, the most affordable, and the most trustworthy. China is betting on open-source and price. America is betting on polish and safety. You, standing in the middle with access to both, are the real winner. Use them all. Compare them. Pick what works. That is how you learn.
Swati Gupta
ByteBuilders
P.S. Which AI model are you switching to this week? Hit reply and tell me. I read every response.