In the Middle East, the indigenous people have developed a "pure-blooded" model, and the MiniMax is

  It's the first time in the history of the world.

Get out of here, folks!

China's big model, out to sea!

For the past two days, a giant pigeon has shocked everyone.

Recently, a group of Saudi netizens announced that we have launched the HUMAIN M3, a 100% Saudi national model.

It surpasses the world's most powerful AI model, fits perfectly into Arabic culture and language, deploys on Saudi domestic servers, and is about to open weight!

The company's background is deep, behind the Saudi Public Investment Fund PIF, and the chairman is the Crown Prince of Saudi Arabia himself.

Well, guys, the Saudi netizens have gone straight to heaven, and the comment section is boiling.

Before the release, researchers from several of X's previous studies found the white spot.

The mixed-expert model of 428 billion total parameters, 23 billion active parameters... how do you get more familiar with this formula?

Soon, God responded: "100% Saudi, aren't you afraid that God will punish you?"

The publisher couldn't help but clarify in person: the underlying core of the HUMAIN M3 is based on the Chinese AI company MiniMax's M3 model.

Today I saw the official press release that just came out.It turns out that HUMAIN used to be China's MiniMax.Strategic partners！

It's not a civilian suit.Saudi Arabia's sovereign wealth fund (SAF) is a state-owned enterprise.AIOn the doorstep, real gold and silver cashed checks and became big business customers of China's big model!

It's a country-level doorway built by the Turks of the Middle East in gold, and it's a "Chinese heart".

China's big model is now beginning to become someone else's "national foundation".

Maybe you'll think this isn't a setup.

It's too small a Saudi ambition to think about that.

The Arabic language is one of the most complicated languages in the world.

It is right-to-left slanted, complex in form, and heavily differentiated in dialects, such as Egyptian Arabic, Gulf Arabic, and Levantine Arabic, just like several different languages.

I greeted the Arabic in the dining room of the Dragon Restaurant and felt closer to spreading the wider Egyptian Arabic dialect.

Moreover, high-quality Arabic content has long been scarce on the Internet.

If you were to deal with Arabic using the general model of the United States, you'd always have a hot air balloon.

What did the Saudis do?

They took the powerful MiniMax M3 backbone directly and "throw" more than 1 trillion pure-Arabic Tokens into it after training.

This trillion-dollar resource is complemented not only by language, but also by local dialects, indigenous cultures, religious practices and even the way of thinking of the Arab region.

The result was a powerful miracle.

In seven Arabic-language benchmark tests, HUMAIN M3 scored the highest five points, with an equality average of up to 89.37%.

This score directly rubs the GPT-5.6 Sol and the Claude Opus 5 top-of-the-line closed-source models against the ground.

In the Arabic-language vertical race course, it is up 9 percentage points on average from the original base, achieving a real lead.

This is no longer a model that speaks Arabic, but a model created specifically for Arabic.

And, unfortunately, it's born with an innate multi-mode capability to understand images, long videos, and even interact directly with computer screens, all for the future workflow of AI.

In fact, this collaboration is significant because it represents a new form of Chinese AI outsourcing.

In the past, China's AI went overseas, either by bringing apps to overseas purchases or by selling APIs to overseas developers based on the number of calls made.

Although money can be made quickly, fate is always in the hands of others and can be changed at any time.

But this time, the collaboration between MiniMax and Saudi Arabia is an entirely new game: the Chinese model is becoming the technological backbone of other countries' AI systems!

It's not uncommon for people to think of Huawei back then.

Previously, Chinese mobile phone manufacturers went overseas to sell a few cheap phones. But Huawei went overseas to help carriers in the Middle East, Africa, and Europe build 4G, 5G communication bases and lay submarine cables.

Selling phones is business, building a base station is ecology, making rules.

Today's MiniMax is about building an AI-era overseas base.

Perhaps in the future, many other countries will have their own big models, underpinned by MiniMax, Qwen, DeepSeek, GLM. This is the real outward spread of the Chinese open source model.

What's exciting is that the HUMAIN platform will become Saudi Arabia's national AI infrastructure connecting Saudi developers, researchers and businesses in the future.

This means that the entire future of AI in Saudi Arabia, including autonomous driving, government systems, and medical AI, is likely to be built on top of this Chinese base.

This is arguably the first time that China's big model has really killed the core of global sovereign AI.

Why would Saudi Arabia spend so much money instead of creating its own "sovereign grand model" or even inviting the crown prince to sit in town?

Because in the age of AI, data is oil and models are refineries.

As a powerhouse in the Middle East that relies heavily on energy transformation, it is impossible for Saudi Arabia to transmit all of its core data - political, financial, medical, and cultural - to the servers of OpenAI or Anthropic in California, USA.

If one day, a ban sheet is sealed, the entire country's AI ecosystem will collapse.

As a result, "sovereign AI" has become a necessity in non-English speaking countries around the world.

In order to do this, large models must meet these conditions of deployment on their own servers, having local cultural values, and being fully autonomously controlled.

At that time, China's open source model became the world's best ally.

Hugging Face's latest report shows that in the past year, China's R&D models accounted for 41% of platform downloads, surpassing the United States for the first time in history!

Silicon Valley is trying to do "one of the strongest models in the world".

And China's big model, in addition to being the world's strongest model, is going the other way: we provide the best open source base, you take it, combine your language, your culture, your data, build your own model.

If the base model itself does not have SOTA-level hardness, any post-training will be difficult to achieve.

It turns out that for the non-English speaking world, the paradigm of "strong core+localized deep post-training" is far better than a general closed-source model based on English.

China's open-source database + local data fine-tuning is the most perfect route.

The HUMAIN M3 has proven this by outperforming GPT and Claude in various assessments.

Interestingly, if you look at the recent big national moves, you will find that this collaboration is by no means a coincidence.

In a recent WAIC high-level speech, it was clear that "supporting global AI capacity building" and even specifically "building an international AI application cooperation center for the UAE" were needed.

The birth of the HUMAIN M3 coincides well with this declaration.

Driven by "Vision 2030", Saudi Arabia is betting on AI and is reportedly planning to invest up to $100 billion to build a national-level AI giant.

There is no shortage of money in Saudi Arabia, but the "digital colonization" that Saudi Arabia is worried about, and the US giant that can't solve the problem of US AI models such as Apple Fable, is constantly facing scrutiny and disruption.

Even more deadly, core data, such as government and energy in the Middle East, must be uploaded to the cloud to directly trigger local bill redlines.

And China's big model is Saudi Arabia's most urgently needed "sense of digital security".

Europeans and Americans want to sell you leased software, but China is willing to help them build a "blood-making system" in their own cabin in Saudi Arabia.

Saudi Arabia has money, computing power, massive amounts of Arabic-language data, China has a top-notch basic model architecture like the 428B MoE, mature training engineering experience, and a more open and inclusive eco-sharing spirit.

Two ancient civilizations are intertwined in the age of AI.

China's large-scale model, streamlined as "hydroelectric coal", enters the servers of Middle Eastern indigenous peoples and fosters a vibrant Arabic-language intelligent ecosystem.

Looking back over the past few decades, we seem to have always been followers.

In the PC era, the bottom line is Windows, we make software; in the mobile era, the bottom line is iOS and Android, we make apps.

For a long time, even the most profitable internet companies in China have been living by the rules of others.

But today, the script was finally rewritten.

China's big model has begun to become the "backbone of the nation" for others.

From the MiniMax M3 to the HUMAIN M3, the legacy of technology spans thousands of miles. In the global ecosystem of open-source AI, we have climbed to the top of the food chain.

Today, hundreds of millions of people around the world speak Arabic, and hundreds of millions more in the future will speak Hindi, Spanish and Swahili.

They will find that the most open, best-used basics will be engraved with the words AI, made in China.

China's open-source model of the Star-Spangled Banner is just getting started.