Your AI Roundup
Key developments across open-source AI, investment, and emerging startups
Welcome to this week’s AI round-up! This edition examines how open-source AI continues to evolve, from language diversity to adoption trends. We also review last week’s investment activity and showcase startups that are advancing AI at the intersection of software, chemistry, and grid reliability.
This week at a glance:
The growing number of non-English models: our analysis of Hugging Face data shows accelerating growth beyond English, pointing to a more geographically distributed open-source AI ecosystem.
Chinese developers drive adoption: over 45% of top open-model public downloads on the Hugging Face Open Model Leaderboard come from Chinese developers, with video and visual creation tools leading engagement beyond text.
2025 has been the most eventful year for open source AI: a breakthrough year for open source AI, with China leading model development, major advances in multimodality and reasoning, and growing recognition in policy and infrastructure initiatives.
Weekly funding: Education leads the week, with companies in the sector raising more than $1 billion.
Research papers: five notable publications to keep an eye on this week.
Startup spotlights: three companies applying AI across API security, molecular discovery, and grid reliability.
1. The growing number of non-english models
Analysis of Hugging Face data indicates that open-source AI development is becoming more multilingual and geographically diverse. Although English remains dominant in absolute terms, Ukrainian, Swedish, Arabic, Turkish and Chinese languages experienced the fastest year-on-year growth between 2024 and 2025.
These higher growth rates suggest that participation is expanding from previously underrepresented linguistic communities, indicating a shift towards a more diverse and globally rooted open-source AI ecosystem.
→ Explore the interactive chart for a detailed view
2. Chinese developers account for over 45% of top open-model public downloads
Data from Hugging Face shows that, in 2025, over 45% of downloads of the most popular open-source models came from Chinese developers.
This year, Qwen2.5 models were downloaded over 750 million times, while Meta’s LLaMA 3.1 and 3.2 models surpassed 500 million downloads despite being released in 2024.
Beyond text models, video and visual creation tools (e.g. ByteDance’s AnimateDiff-Lightning and the Comfy/ComfyUI ecosystem, as well as releases from Stable Diffusion and Black Forest Labs) are driving adoption across modalities. Community contributions are also expanding the ecosystem, demonstrating that both developer-led and user-driven activities influence the global distribution of open-source AI usage.
→ Explore the interactive chart for a detailed view
3. 2025 has been the most eventful year for open source AI
2025 was a pivotal year for open-source AI. Following the release of DeepSeek-R1, momentum accelerated as advancements in mixture-of-experts architectures drove rapid progress in reasoning and multimodal capabilities.
Model families such as Alibaba’s Qwen, Moonshot AI’s Kimi-K2, Zhipu’s GLM-4.5 and MiniMax’s M2 have cemented China’s position as a leading contributor to open models. Meanwhile, fewer but notable releases have continued to emerge from the US and Europe.
Beyond open weights, initiatives such as Switzerland’s Apertus, the Allen Institute’s OLMo-3 and Molmo-2, and NVIDIA’s Nemotron-3 have increased transparency surrounding training processes.
4. When AI misses the holiday moment
In early December 2025, McDonald’s Netherlands released a Christmas advert largely created using generative AI. The advert combined AI-generated imagery with a parody of a traditional festive song. Following widespread criticism of its visuals and perceived lack of emotional warmth, the advert was removed from the company’s official YouTube channel within days.
This echoes similar responses to recent AI-driven campaigns by brands such as Coca-Cola and Valentino, where audiences have questioned whether the creative output aligns with established brand standards. These cases highlight the ongoing challenges of using AI for sentiment-heavy advertising, where stylistic inconsistencies and concerns about creative labour can undermine audience engagement and brand perception.
5. AI investments of the week - Week #51
AI companies raised $4.33 billion globally this week, reflecting continued capital deployment across a wide range of AI applications. The largest rounds 🇺🇸 OpenAI $1 billion, 🇸🇦 Noon $500 million, 🇬🇧 PolyAI $86 million, 🇧🇷 Morada.ai $3.13 million.
Most active sectors in AI investment this week:
Education: OpenAI (US), PolyAI (GB), RelationalAI (US), Oboe (US), Typecast (aka Neosapience) (KR), BoodleBox (US), Tunib (KR), Livetoon (JP)
Industry & Mobility: Unconventional AI (US), Generative Bionics (IT), Deep Robotics (CN), Medra AI (US), Bobyard (US), Assaia (CH), Xingyuanzhi Robotics (CN), Matta (GB), OTIV (BE), Mach (US), Ulink Intelligent (CN), Medida AI (US), Fly Media (US), Neurophos (US), GRACIA.AI (US), Zentio (DE), Chemify (GB), Buildroid.ai (US)
Software: Harness (US), Fal (US), Port (IL), Serval (US), Runware (GB), Solve Intelligence (US), Hud (US), Prime Security (US), Notta (JP), Equixly API Security (IT), EVERTRUST (FR), Vybe (US), Spoor (NO), Worktrace AI (US), Kilo Code (US), Buildcheck (US), Cleric (US), Ageiro Ltd (GB), Empromptu (US), Recho (JP), Entrokey Labs (US), Zof AI (US), Intellicule (US)
→ Explore the interactive chart for a detailed view of trends
6. This week’s most notable AI research papers
This week, research highlights the future of AI at the edge, workforce transformation, economic inclusion, information integrity, and global governance.
Curious for more? Explore this week’s 5 most notable AI research papers to stay up to date on the latest insights and developments in the field.
7. Startup spotlights of the week
This week’s startup selection highlights how AI is being applied to critical infrastructure challenges across software security, energy systems and molecular discovery.
🇮🇹 Equixly: develops AI agents that continuously assess API security during development by analysing business logic, data flows and attack paths. The Florence-based company recently raised a €10 million Series A to scale its platform and expand internationally, starting with the UK.
🇸🇪 Elvy: offers an AI-driven home energy subscription service that optimises solar, storage and heating systems, reducing reliance on the grid while stabilising household costs. The Stockholm-based start-up recently secured €500 million in funding to increase its presence in Sweden and prepare for a European rollout.
🇬🇧 Chemify: combines AI, robotics and programmable chemistry to automate the design and synthesis of complex molecules for use in pharmaceuticals and materials science. This University of Glasgow spin-out company received $1.6 million from the Bill & Melinda Gates Foundation to speed up the discovery of drugs for tuberculosis in collaboration with Lgenia.
What’s next
For more updates, research, and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s roundup directly in your inbox and stay ahead of key developments.
Have a great week ahead!
Gaia Cavaglioni
On behalf of the AI World Team