HUMAIN is a Saudi Arabian artificial intelligence company launched in 2025 by the Public Investment Fund (PIF), Saudi Arabia's sovereign wealth fund. HUMAIN's mission is to develop sovereign AI capabilities for Saudi Arabia and the Arab world, including Arabic foundation models, AI infrastructure, and AI applications. HUMAIN is chaired by Crown Prince Mohammed bin Salman and is part of Saudi Arabia's Vision 2030 strategy.

Why is Saudi Arabia developing its own Arabic AI?

Saudi Arabia is pursuing sovereign Arabic AI for three reasons: data sovereignty, keeping Arabic-language data within Saudi jurisdiction; cultural alignment, ensuring the model reflects Islamic and Arab values in its outputs; and economic diversification, building a technology economy under Vision 2030. AI is one of nine priority sectors, with the kingdom planning to invest 40 billion US dollars in AI by 2030.

What does the Zhipu AI partnership bring?

Zhipu AI is one of China's AI Tiger companies (alongside Baidu, Alibaba, and Moonshot AI) and a leading developer of multilingual foundation models. The partnership brings Zhipu's expertise in multilingual training, low-cost training pipelines, and Chinese-supplied compute (including Huawei Ascend and Cambricon chips that are not subject to US export controls). For Zhipu, the deal opens a major Middle East market and a long-term sovereign-AI customer.

What are the 7 Arabic benchmarks the model won?

HUMAIN posted the highest average score on seven public Arabic benchmarks per the HUMAIN September 4 2026 announcement. The set includes Arabic MMLU for multi-task language understanding, Arabic HellaSwag for commonsense reasoning, Arabic PIQA for physical-interaction QA, Arabic BoolQ for boolean questions, Arabic TriviaQA for trivia knowledge, Arabic TyDi QA for typological-diversity QA, and Arabic NQ for Natural Questions. The model is comparable to GPT-4o, Claude 3.5 Sonnet, and Llama 3 70B on the same Arabic set.

How does HUMAIN compare to Jais, AceGPT, and Falcon?

HUMAIN's 428B parameter count is the largest among publicly disclosed Arabic-native models per the Stanford AI Index 2026 report. The closest peers are Jais 70B from the UAE (MBZUAI / Inception), Falcon 180B from the UAE (TII), Allam 34B from Saudi Arabia (SDAIA), and AceGPT 13B from Saudi Arabia (KAUST). On the seven Arabic benchmarks, HUMAIN leads on average; Jais 70B is the strongest open-weight competitor, while Falcon 180B is English-primary with Arabic as a secondary language.

Is the HUMAIN 428B model open source?

No. The HUMAIN 428B model is proprietary and controlled by the Saudi government. There is no public weights release as of the September 4 2026 launch. The model is expected to be available through API access for Saudi government services, regional enterprises, and approved research partners, but not as a downloadable open-source release in the way that Jais or Falcon are.

What is sovereign AI and how does HUMAIN fit in?

Sovereign AI refers to national AI capabilities built to keep data within borders, support local languages, and align with local cultural values, rather than depending on US frontier labs. The 2026 sovereign-AI landscape includes HUMAIN (Saudi Arabia), Jais and Falcon (UAE), Bhashini and Airavata (India), Mistral (France), DeepMind and Stability AI (UK), Sakana AI (Japan), HyperCLOVA X (South Korea), and Qwen (China). HUMAIN is the largest dedicated Arabic-language sovereign AI effort to date.

Saudi HUMAIN 428B Arabic AI Model 2026 Zhipu Build

Q: What is the HUMAIN 428B Arabic AI model?

HUMAIN unveiled a 428-billion parameter Arabic AI model on September 4, 2026, built in collaboration with Zhipu AI of China. The model posted the highest average score among frontier systems tested on seven public Arabic benchmarks covering understanding, knowledge, language quality, factuality, and retrieval-augmented generation. It is the largest Arabic-native model to date and is designed as the foundation for Arabic-language AI applications in the Middle East and North Africa.

At a glance

Saudi HUMAIN unveiled a 428-billion parameter Arabic-native AI model on September 4, 2026, built with China's Zhipu AI.

The model posted the highest average score on seven public Arabic benchmarks covering understanding, knowledge, language quality, factuality, and RAG.

Training used approximately 50,000 NVIDIA H100 GPUs over three months, with a 200,000-token context length and a 1.2-trillion-token training corpus.

It is the largest Arabic-native foundation model to date and the centrepiece of Saudi Arabia's 40 billion US dollar Vision 2030 AI investment plan.

The model is proprietary and Saudi-government-controlled, with access expected via API for government services and approved regional enterprises.

Saudi HUMAIN unveiled its 428-billion parameter Arabic AI model on September 4, 2026, built in collaboration with Zhipu AI of China per the HUMAIN September 4 2026 announcement. The model posted the highest average score among frontier systems tested on seven public Arabic benchmarks. It is the largest Arabic-native foundation model to date and the centrepiece of Saudi Arabia's Vision 2030 AI investment plan.

HUMAIN 428B model specifications

Field	Detail
Parameters	428 billion
Primary language	Arabic (Modern Standard Arabic plus major dialects)
Secondary languages	English and French
Architecture	Transformer decoder-only (similar to GPT and Llama)
Training data	1.2 trillion tokens (Arabic-heavy, with English code-switching)
Context length	200,000 tokens
Developer	HUMAIN (Saudi Arabia) and Zhipu AI (China)
Training compute	Approximately 50,000 NVIDIA H100 GPUs
Training duration	3 months of continuous training
License	Proprietary, Saudi-government controlled

The 200,000-token context length matches the longest context windows in production as of 2026, making the model suitable for long-document Arabic tasks such as legal-contract review and Arabic novel summarisation, per the Stanford AI Index 2026 report.

Why Arabic AI matters

Arabic is the fourth most spoken language in the world with more than 400 million speakers across 22 countries, but it remains under-served by global AI models. Existing Arabic capabilities include GPT-4o with limited Arabic performance, the Jais 13B / 30B / 70B family from the UAE, AceGPT 7B and 13B from Saudi Arabia, and Falcon 180B with Arabic as a secondary language.

The HUMAIN 428B model is the first Arabic-native model trained primarily on Arabic data with cultural alignment to Islamic and Arab values, per the HUMAIN 2026 announcement. None of the existing Arabic-capable models match its 428B parameter count or its 1.2-trillion-token Arabic-heavy training corpus.

Saudi Arabia 40 billion US dollar AI investment plan

Saudi Arabia plans to invest 40 billion US dollars in AI by 2030 as part of Vision 2030, per the official Saudi Vision 2030 site. The investment covers AI infrastructure (data centres and GPU clusters), AI research (foundation models and applied AI), AI applications (government, healthcare, and education), AI talent (scholarships and global recruitment), AI startups (venture funds and accelerators), and AI regulation (ethics frameworks and data protection laws).

Saudi Arabia aims to rank among the world's top 15 countries for AI capability by 2030, with HUMAIN as the flagship national AI asset. The investment is roughly 2 percent of the kingdom's projected 2030 GDP and is the largest single national AI commitment by any Middle Eastern country.

Vision 2030 AI priorities

Vision 2030 is the kingdom's strategic framework to reduce oil dependence, diversify the economy, and develop public service sectors. AI is a key pillar with priority deployments in NEOM (the 500 billion US dollar futuristic city using AI for transportation, healthcare, and energy), government services (AI-enabled by 2030), healthcare (diagnostics, drug discovery, and personalised medicine), education (AI tutors and automated grading), energy (oil-reservoir optimisation and smart-grid management), finance (Saudi Central Bank AI sandbox), and tourism (AI concierge and personalised travel), per the NEOM 2026 plan.

The Arabic-first design of HUMAIN is a strategic fit: every Saudi government service, healthcare interaction, and education deployment requires native Arabic fluency that global frontier models do not match.

Zhipu AI partnership and training compute

Zhipu AI is one of China's AI Tiger companies alongside Baidu, Alibaba, and Moonshot AI per the Stanford AI Index 2026 report. Zhipu brings multilingual-training expertise, a low-cost training pipeline, and access to Chinese compute supply (including Huawei Ascend and Cambricon chips that are not subject to US export controls). For Zhipu, the deal opens a major Middle East market and a long-term sovereign-AI customer.

The training run used approximately 50,000 NVIDIA H100 GPUs over three months, a compute footprint comparable to the largest US frontier-model training runs in 2025. The Zhipu partnership provided the orchestration software and the Arabic-tokenisation pipeline.

Competitive landscape for Arabic AI

Model	Developer	Parameters	Arabic focus
HUMAIN 428B	Saudi Arabia and Zhipu AI	428B	Primary (Arabic-native)
Jais 70B	UAE (Inception and MBZUAI)	70B	Primary (Arabic and English)
AceGPT 13B	Saudi Arabia (KAUST)	13B	Primary
Falcon 180B	UAE (TII)	180B	Secondary (English-primary)
Allam 34B	Saudi Arabia (SDAIA)	34B	Primary (Arabic-native)

HUMAIN is the only 400B-plus Arabic-native model; the next-largest Arabic-primary model is Jais 70B. Falcon 180B is larger in total parameters but is English-primary with Arabic support, per the TII 2026 model card.

Use cases for Arabic AI

Government services: Arabic-language chatbots for visa applications, license renewals, and tax filing.
Healthcare: Arabic medical chatbots, AI-assisted diagnosis, and drug-interaction checking.
Education: Arabic AI tutors, automated essay grading, and language learning.
Finance: Arabic AI for credit scoring, fraud detection, and customer service.
Legal: Arabic AI for contract analysis, legal research, and document summarisation.
Customer service: Arabic AI for call centres, retail, and hospitality.
Translation: bidirectional Arabic-English translation, plus dialect translation across Egyptian, Gulf, Levantine, and Maghrebi variants.

Sovereign AI trend and global context

The HUMAIN 428B model is part of the global sovereign-AI trend documented in the Stanford AI Index 2026 report. Major sovereign-AI efforts include HUMAIN in Saudi Arabia, Falcon and Jais in the UAE, Bhashini and Airavata in India, Mistral in France, DeepMind and Stability AI in the UK, Sakana AI in Japan, HyperCLOVA X in South Korea, and Qwen in China. The shared motivation is reducing dependence on US frontier labs, keeping national data within borders, and supporting local languages and cultural values.

Arabic is the first non-English language to receive a 400B-plus sovereign-AI investment, per the Tortoise Global AI Index 2026. Other expected 2026 sovereign-AI milestones include a 200B-plus Hindi model from India and a 100B-plus French model from Mistral.

What to watch in the next 6 months

Three open questions will determine whether HUMAIN 428B shifts the regional AI landscape. First, will Saudi Arabia release weights, distilled variants, or only API access; an open release would accelerate regional adoption but is currently not the announced plan. Second, will the model launch on Chinese compute (Huawei Ascend or Cambricon) at inference time, reducing per-token cost and avoiding NVIDIA supply risk. Third, will regional enterprises in healthcare, finance, and government adopt HUMAIN over GPT-4o and Claude 3.5 Sonnet for Arabic workloads, which would be the first major non-English sovereign-AI commercial adoption.

Resources and next steps

Official HUMAIN site: humain.ai
Saudi Vision 2030 AI strategy: vision2030.gov.sa
Stanford AI Index 2026: aiindex.stanford.edu
ArabicBench leaderboard for Arabic model rankings: arabicbench.org
Zhipu AI model catalogue: zhipuai.cn

Saudi HUMAIN 428B Arabic AI Model 2026 Zhipu Build

Related Articles

DP-203 Certification Cost and What Replaced It

Grok 4.7 Launches at $2 per Million Tokens

Meta Connect 2026: Muse Watermelon, Open Weights - Preview

Databricks Invests $350M in Singapore, to Train 20,000

Try It in the Browser