Humain, an artificial intelligence company backed by Saudi Arabia’s sovereign wealth fund PIF, has released a new AI model called humain-m3 based on China’s MiniMax M3. The company said the model is designed to strengthen the Arabic-language AI ecosystem and support the development of frontier artificial intelligence. The launch also points to a wider push by more countries to gain greater control over critical AI technologies through domestic AI platforms and open-source models. Earlier, Humain said it planned to raise its first $2.5 billion for a fund focused on data center investments. According to Bloomberg, people familiar with the matter said the fund would help finance 250 megawatts of data center capacity being developed by Humain in partnership with Al Moammar Information Systems, with the broader buildout potentially expanding to 1 gigawatt over time.

Saudi Arabian AI company HUMAIN launched its M3 model, initially claiming it as '100% Saudi'. After scrutiny, it clarified that the model is built upon China's MiniMax M3. Both use a Mixture-of-Experts architecture with 428 billion total parameters and 23 billion active parameters. HUMAIN performed post-training on over 1 trillion Arabic tokens, achieving an average score of 89.37% on 7 Arabic benchmarks. The company plans to open-source weights and deploy on Saudi servers. Analysts say this shows Chinese models are becoming the technical foundation for sovereign AI in other countries.

Hong Kong-listed MiniMax Group Inc. (stock code: 0100) has released its unaudited interim results for the six months ended June 30, 2026, showing sharp revenue growth alongside continued losses and heavy research spending. Total revenue reached $116.6 million, up 283.1% year over year, while gross profit rose 464.8% to $20.813 million. Loss for the period came in at $358 million, narrowing 11.0% from a year earlier, but adjusted net loss widened 111.2% to $293 million. By segment, revenue from the open platform and other AI-based enterprise services climbed 703.1% to $73.929 million, accounting for 63.4% of total revenue, while AI-native product revenue increased 100.9% to $42.644 million. Mainland China contributed 39.2% of revenue, with other regions making up 60.8%. Research and development expenses rose 138.8% to $296.9 million. The company said its products and services now cover more than 230 countries and regions, and noted that it launched MiniMax M3, upgraded its core model products, and later released MiniMax H3 with open weights during the reporting period.

OpenRouter data shows Chinese open-weight models now account for 61% of platform token consumption, with DeepSeek and GLM nearing US frontier performance. The gap holds at 3-6 months and is not widening.

MiniMax open-sourced its flagship M3 model today on Hugging Face. It features a 428B MoE architecture with only 23B active parameters per token, supports 1M token context, and achieves up to 15x decoding speedup via its proprietary MSA sparse attention mechanism.

MiniMax used the 2026 World Artificial Intelligence Conference, or WAIC 2026, to present its flagship MiniMax M3 model alongside a broader lineup of multimodal models and AI-native products. The company also gave an early preview of H3, its new-generation multimodal generation model, during the event. According to The Paper, the showcase was built around ecosystem collaboration cases spanning robot dogs, AI glasses, smart earbuds, NAS, financial services and film and television content. Through those examples, MiniMax outlined how its work extends from foundation models and agent products to concrete applications. The presentation covered model capabilities, academic research, product tools, multimodal creation and industry use cases. Visitors were able to review M3’s capabilities in long-context processing, coding and agentic tasks, while also seeing how multimodal models are being applied in creative workflows, smart hardware and industry services. MiniMax also shared updates on its open-source ecosystem, industry partnerships and global expansion efforts.

Goldman Sachs says China’s large language model market is moving into a decisive stage, with open-source and open-weight models closing the gap with top proprietary systems while scaling adoption among domestic companies and overseas small and medium-sized businesses. In a 50-page report led by analyst Ronald Keung, the bank examines how Chinese model developers are delivering high performance at lower cost, why they chose open deployment paths, where monetization may improve, and which companies are best positioned for the long run. The report identifies Zhipu and DeepSeek as the strongest names in foundational text models, while ByteDance leads in multimodal applications. Goldman also keeps Buy ratings on MiniMax and Kuaishou. It argues that architecture choices such as mixture-of-experts and sparse attention have helped Chinese developers cut training and inference costs even with smaller parameter footprints and limited access to advanced chips. Goldman estimates China’s AI model API and subscription revenue could rise from RMB 35 billion in 2026 to RMB 879 billion in 2030, with daily token consumption increasing from 350 trillion to 4.6 quadrillion over the same period. The bank also expects the market to shift from fully open MIT-style licensing toward open weights with community licenses and revenue-sharing arrangements, a move it says could materially improve unit economics for model providers.

B.AI and MiniMax have teamed up to offer a limited-time free trial of the M3 model, leveraging full-stack infrastructure to lower the entry barrier for Agentic productivity and achieve zero-cost deployment.