HUMAIN AI Stack December 2026 Explained: 3 Trillion Parameter Model, 15 December Date and What Is Still Unproven

Majdoul Tower in Riyadh, where HUMAIN's offices are and where CEO Tareq Amin announced the December AI stack on 11 November 2026. Image: Wikimedia Commons / Kolaiel, CC0.

TL;DR: HUMAIN's AI stack December 2026 explained: Saudi AI company HUMAIN says its full AI stack, built around a 3 trillion parameter open-weight model its CEO calls the world's largest, arrives on 15 December 2026, with its HUMAIN+ learning platform following in Q1 2027. The date is firm; the model's name, benchmarks and price are not.

Key Highlights

• Full stack date: 15 December 2026, announced by CEO Tareq Amin on 11 November at HUMAIN's offices in Majdoul Tower, Riyadh.
• The model: 3 trillion parameters, described by Amin as the world's largest open-weight model. No name, benchmarks or price were disclosed.
• The cost claim: 88% lower cost and token consumption than models available outside Saudi Arabia, with no comparison models named.
• HUMAIN+: the applied AI learning platform gets an operational launch in Q1 2027, with a Riyadh campus for about 1,500 learners a year, built with VUZ.
• The catch: nothing about the model has been independently verified yet, availability outside Saudi Arabia is unclear, and HUMAIN's previous flagship, humain-m3, was built on China's MiniMax M3.

HUMAIN AI stack December 2026 explained: what was announced

On Wednesday evening, 11 November 2026, HUMAIN chief executive Tareq Amin told press, AI experts and prospective HUMAIN+ participants gathered at the company's offices in Majdoul Tower, Riyadh, that HUMAIN's complete AI stack would be ready in December. His words, reported by Arab News: "By December, mark my words ... the world's largest open-weight model, 3 trillion parameters model, will be available in Saudi (Arabia), nobody in the planet has this." Then the date: "The complete stack, December 15, is not far away."

HUMAIN is the PIF-owned Saudi AI company set up in May 2025 to cover data centres, cloud platforms and AI models. The pitch Amin made is sovereignty in its fullest form: "Sovereign infrastructure, deployed here, ready to be consumed by any builder, enterprise, government without any control claim that leaves the country." The software stack is built by Saudis, the cloud inferencing is built by Saudis, and the infrastructure is deployed and operated by Saudis, he said, adding that "for the first time, HUMAIN owns the entire stack." He said the work took 15 months, calling it "nothing but a miracle of what this team has done."

The framing matters as much as the date. Amin's stated obsession is affordable computing power for entrepreneurs, startups and students, and he paired that with a second promise: "a state-of-the-art model that is 88 percent less cost and token consumptions than available models outside of the country." The stack is meant to be something developers build on directly: "I want to offer the entire stack of HUMAIN that HUMAIN+ sits on top of, low-cost model, state-of-the-art frontier model, so people could actually leverage HUMAIN code to build their products, platforms and application."

That is the announcement. What follows is the part that decides whether 15 December is a milestone or a press line: everything about the model itself is, for now, a claim made on stage, not a demonstrated fact.

The 3 trillion parameter model: what we know and what we do not

The single biggest number in the announcement is also the least documented. Amin said the model has 3 trillion parameters and is the world's largest open-weight model. He did not name the model, show any benchmark, publish any weights, or give a price or an access route. Arab News, which reported the event, did not independently check the claims, and no other outlet has been able to either, because there is nothing public to check yet.

For scale, 3 trillion parameters would be roughly seven times the 428 billion parameters of humain-m3, the Arabic model HUMAIN unveiled at LEAP 2026 in September. But that comparison carries a warning of its own: humain-m3 was not trained from scratch by HUMAIN. It was built on MiniMax M3, the open-weight model from Shanghai-based MiniMax, with HUMAIN's contribution being further pre-training on more than one trillion tokens of Arabic-native content. HUMAIN reported an average of 89.37% across seven Arabic benchmarks for humain-m3, but those scores were measured on its own infrastructure and the model was never submitted to the Open Arabic LLM Leaderboard, so no independent ranking exists.

That history does not disprove the new claim. It does set the test for 15 December: whether HUMAIN publishes the model's weights, names its architecture and training provenance, and releases performance figures that anyone else can reproduce. Until then, "world's largest" is the CEO's description, not an established fact, and this article treats it that way throughout.

Claim or fact	Status on 13 November 2026
Full AI stack available 15 December 2026	Announced by the CEO on the record, 11 November
3 trillion parameter open-weight model	Claimed by Amin; model unnamed, no weights or benchmarks shown
"World's largest open-weight model"	CEO's claim; not independently verified
88% lower cost and token use	CEO's claim; no comparison models named
HUMAIN+ operational launch, Q1 2027	Announced 11 November; PR release of 11 November said "next few months"
Stack available outside Saudi Arabia	Unclear; no price or access route given

Illustrative image: a data centre. HUMAIN's stack pairs its model with Saudi-built cloud and infrastructure. Image: Wikimedia Commons / BalticServers, CC BY-SA 3.0.

HUMAIN+ moves to the first quarter of 2027

The second half of the announcement is the learning platform, HUMAIN+, and here the timetable actually firmed up. When HUMAIN and VUZ announced their partnership on 11 November, the platform was due "in the next few months". A day later, Amin put a quarter on it: operational launch in the first quarter of 2027, sitting on top of the December stack.

HUMAIN+ is an applied AI education platform, built with VUZ, a Riyadh-founded company specialising in patented 360-degree video technology. It combines HUMAIN Voice and HUMAIN Create, and it is designed around role-based learning, sector-specific labs, hands-on product development and competency-based assessments, rather than traditional courses. There are four learning journeys, for leadership, non-technical professionals, technical talent and students, covering AI strategy, governance and productivity, product development, agentic AI, security and infrastructure operations. The sector labs span government, finance, healthcare, energy and industry, education, tourism, logistics and supply chain, legal and compliance, human resources and sport.

The Riyadh campus is designed for about 1,500 learner places a year, supported by GPU-backed Builder Labs and spaces for demonstrations and project showcases; a ribbon-cutting at the event inaugurated the learning centre. Amin's framing for the platform is deliberately grand: dissatisfied with existing training methods a year ago, he asked, "What if we create the Netflix of AI?" VUZ founder and CEO Khaled Zaatarah called it a long-term partnership to upskill millions of people in the region and globally. Neither report gives sign-up details or pricing for HUMAIN+ yet.

Why it matters: the sovereign AI race, and the India read-across

HUMAIN's announcement lands in a week when the whole AI industry has turned to cost and residency. Google launched Gemini 4 Hydrogen, its fast $1.50 model, on 12 November, and OpenAI opened local processing for its API in India the same day. The competition is no longer only about who has the smartest model; it is about who can run AI cheaply, close to the user, under local rules. HUMAIN's answer is the most absolute version of that idea: an entire stack, from data centre to model, inside one country, under one owner.

That is also why the December test matters beyond Saudi Arabia. If the weights really are released and the 88% cost claim holds up against named models, HUMAIN becomes a genuine alternative source of frontier-scale open weights, and a template other governments will study. If 15 December brings a name and a demo but no weights, no benchmarks and no pricing, the announcement will read as a sovereignty statement rather than a product launch. The wider Saudi push gives the claim some scaffolding: AWS says its first Saudi cloud region launches in December 2026 alongside a planned 50MW AI Zone with HUMAIN by 2028, and AMD and HUMAIN already sell "AI in a Box", a turnkey compute package, to Saudi enterprises. But scaffolding is not the model itself.

For India, nothing was announced: no India availability, no India pricing, no local partnership. The read-across is the idea, not the product. India is building its own sovereign-AI capacity through the IndiaAI Mission, and Indian developers and enterprises face the same question HUMAIN claims to answer: how much does it cost to run serious AI, and where does the data sit? If HUMAIN's 88% figure survives contact with reality, every Indian buyer of AI compute will want to know how it was achieved. If it does not, the episode is still useful, as a reminder to read cost claims against named comparisons, not adjectives.

The ownership backdrop is worth remembering too. Aramco closed its 49% stake in HUMAIN on 1 November 2026, paying about $2.6 billion in cash while the PIF kept 51%, with a target of 3 GW of AI data-centre capacity by 2030. The December stack is the first big product test of that ownership: a state-backed company, now part-owned by the world's largest oil producer, promising the world's largest open model in about a month.

Should you care?

Verdict: Watch 15 December, do not plan around it yet. The date is real and the ambition is serious, but a model with no name, no benchmarks, no price and no access route is a promise, not a product. Developers and AI buyers should put 15 December on the calendar and judge HUMAIN on what it publishes that day: weights, architecture, reproducible scores and terms for developers outside Saudi Arabia. If those arrive, this is a major open-weights event. If they do not, treat the "world's largest" line as marketing until proven otherwise.

FAQs

What is HUMAIN's AI stack?
It is HUMAIN's full-stack AI offering: a model, a cloud inference layer and the physical data-centre infrastructure, all built and operated inside Saudi Arabia. CEO Tareq Amin says the complete stack will be available on 15 December 2026, and that for the first time HUMAIN owns the entire stack.

When will HUMAIN's 3 trillion parameter model be available?
According to Amin, the model arrives with the full stack on 15 December 2026. No model name, price or access route has been announced, and it is unclear whether the model and stack will be available outside Saudi Arabia.

Is HUMAIN's model really the world's largest open-weight model?
That is CEO Tareq Amin's claim, not an independently verified fact. He announced a 3 trillion parameter model on 11 November 2026, but gave no name, benchmarks or weights, and Arab News noted the claims could not be independently checked. The test is what HUMAIN publishes on 15 December.

What is HUMAIN+ and when does it launch?
HUMAIN+ is an applied AI education and learning platform built with VUZ, combining immersive video with hands-on labs and four learning journeys for leaders, professionals, technical talent and students. Its operational launch is set for the first quarter of 2027, and its Riyadh campus is designed for about 1,500 learners a year.

Takeaway

HUMAIN has given the AI world a date, a number and a claim: 15 December 2026, 3 trillion parameters, the world's largest open-weight model at 88% lower cost. Two of those three are verifiable in a month, and the third will be judged by them. The honest position until then is to report the promise exactly as it was made, keep the caveats attached, and watch what actually ships. If you want the full ownership picture behind the announcement, read our explainer on Aramco's 49% stake in HUMAIN, and check back after 15 December for what the stack really contains.

Sources

Arab News — HUMAIN CEO unveils December AI stack rollout, HUMAIN+ launch plans for 2027 ( 2026)
PR Newswire — HUMAIN Partners with VUZ to Advance a New Generation of Immersive AI Streaming and Learning by Launching HUMAIN Plus (11 November 2026)
tbreak — HUMAIN's AI stack arrives in December, HUMAIN+ in Q1 2027 (12 November 2026)
Nakoda News — HUMAIN Says Full AI Stack With 3 Trillion-Parameter Model Arrives December 15 (12 November 2026)
vision2030.ai — HUMAIN M3: Saudi Arabia's Arabic AI Model, Examined (September 2026)
HyperFRAME Research — Why Is HUMAIN Building Its Agentic AI Stack on MinIO? (11 September 2026)