open-source AI models

China is increasingly backing open-source artificial intelligence, with Chinese developers and companies releasing AI models whose underlying code is freely available. State media coverage frames the strategy as a way to spread Chinese AI technology globally and compete with closed models from US firms, though no further details were provided in the available coverage.

Nvidia has introduced a new open-source tool designed to improve the security of artificial intelligence systems. The announcement positions the chipmaker as a player in AI safety tooling, alongside its dominant role in AI hardware. Details on the tool's specific capabilities and reception from developers remain limited, but the move aligns with growing industry attention on protecting AI models from misuse and attacks.

Huawei has released openPangu-2.0, open-sourcing the pretraining, supervised fine-tuning and reinforcement learning training code for its Pangu AI models, built to run on its own Ascend chips. The move gives developers full access to the training pipeline and is being read as an effort to build an open ecosystem around Huawei's domestic AI hardware amid competition with Nvidia-based stacks.

Treasury Secretary Scott Bessent said the Trump administration will examine open source AI models, including China's Kimi K3, for potential intellectual property theft, warning that Washington has the ability to sanction those responsible. The remarks come as the Chinese open-source model draws widespread attention for its performance, intensifying debate over US-China AI competition and technology policy.

Stanford University and Nvidia have open-sourced CLM-8B, an AI model built for software agents that caches reusable actions instead of recomputing them. In tests the model ran up to nine times faster than Jev, a comparable agent system. The open release is drawing attention for offering large speed gains on agentic workloads, an area where inference cost is a major bottleneck for developers.

A new open-source tool called Livenerf is drawing attention with a pointed question: has Anthropic's Opus 5.5 model been nerfed? The project aims to monitor whether the AI model's real-world performance has quietly degraded since release, a concern that has grown common among developers who suspect providers downgrade models after launch.

Solus Linux, the independent Linux distribution, has introduced a formal policy governing contributions created with AI and large language models. The move, reported by Linuxiac, sets clear rules for how AI-assisted code is handled in the project. It reflects a broader debate in open-source communities about how to manage the growing volume of AI-generated submissions to volunteer-maintained software projects.

A policy think tank argues the European Union should strengthen its commitment to open-source artificial intelligence, positioning open models as Europe's best route to compete with US and Chinese AI giants. The piece frames open-source AI as a strategic advantage that Europe should build on rather than regulate into weakness.

The open-source project MoneyPrinterTurbo by developer harry0703 is drawing attention for letting users generate high-definition short videos from a single topic or keyword. Built in Python, it combines large AI models with automated workflows to handle scripting, visuals and production with minimal input, reflecting growing interest in fully automated AI video creation tools.

The AI startup VectifyAI has released PageIndex, an open-source Python tool that indexes documents for retrieval-augmented generation without using vector embeddings. Instead of converting text into vectors, it builds a structured, reasoning-based document index that lets large language models navigate documents directly. Developers are taking notice of the unusual approach, which challenges the standard vector-database architecture used in most RAG systems.

Magnitude, a startup in Y Combinator's S25 batch, has launched an open-source self-optimizing inference engine aimed at AI agents. The team says the engine improves how agents run models by adapting and optimizing performance automatically, and has released the code on GitHub. Launch posts of this kind typically draw scrutiny and questions from developers about benchmarks, reliability and how the optimization works in practice.

The Solus Linux distribution has adopted an official policy covering contributions generated with AI tools and large language models. The move sets clear rules for how such code can be submitted to the open-source project. The announcement is circulating in Linux and open-source communities, where projects are increasingly defining their stance on AI-assisted development.

Chinese open-source AI model platforms are competing to establish themselves as China's equivalent of Hugging Face, the leading hub for sharing machine learning models and code. The competition reflects growing momentum in China's domestic AI ecosystem, as developers seek homegrown alternatives for hosting and distributing open models amid a rapidly expanding national AI industry.

A new model focused on cyber open-source intelligence has been released, according to a security researcher announcing it online. The release is being shared among cybersecurity and OSINT practitioners, who are taking note of what a purpose-built model for intelligence gathering could offer. Details on the model's capabilities, creators and intended use were not immediately available.

An expert argues that AI safeguards and regulations effectively build a protective moat around large US AI companies while holding back open-source models from advancing. The claim suggests compliance costs and restrictions favor established firms with resources, disadvantaging smaller developers and the open-source community. The comment adds to ongoing debate over whether AI regulation protects the public or entrenches incumbent tech giants.

Chinese AI firm DeepSeek and Huawei are partnering to build open-source software that lets AI models run on Huawei chips instead of Nvidia hardware. The move aims to reduce China's dependence on US chip technology amid tightening export controls, and signals a push to make domestic AI infrastructure viable at scale.

Chinese AI lab DeepSeek has released open-source software tools that allow its AI models to run on Huawei chips, according to a report by The Information. The move signals deepening efforts to build AI infrastructure without US hardware, and underscores Huawei's growing role in China's semiconductor and AI ecosystem amid export restrictions.

China is holding its World Internet Conference summit with a focus on open-source artificial intelligence, according to CGTN reporting. The event is expected to bring together officials, tech companies and researchers to discuss the development and sharing of AI technologies, reflecting Beijing's push to promote open-source models as part of its broader AI strategy.

DeepSeek has released open-source software enabling its AI models to run on Huawei's Ascend chips, a move read as China's AI sector rallying around domestic semiconductor alternatives to Nvidia. The release highlights growing efforts to build a self-sufficient AI hardware and software stack in China amid ongoing US export restrictions on advanced chips.

A talk from BSides Luxembourg 2026 titled 'Security For AI: AIDR Bastion As Open Source LLM Firewall / AI Prompts Reverse Proxy' by Andrii Bezverkhyi has been made available online, alongside the other talks from the event's track. The presentation covers an open-source firewall and reverse proxy for filtering and securing prompts sent to large language models, part of ongoing cybersecurity discussion about defending AI systems.

Tencent has open-sourced its large Hy4 770B-parameter AI model under the permissive Apache 2.0 license, while also introducing IQuest-Q1, a much smaller 15B-parameter model aimed at more accessible use. Thai-language tech commentary is highlighting the release, noting the combination of a very large open model and a lightweight companion option for developers.

A developer is warning that OpenCode's model whitelist is effectively a list of promises made by other people's APIs, and that the list is slowly rotting as providers change or drop endpoints. A small tool called ocprobe has been introduced to check which models still actually work, highlighting how fragile dependency on third-party APIs can be for open-source coding tools.

Chinese AI firm DeepSeek has released open-source tools designed to work with Huawei's Ascend AI chips. The move supports running AI models on domestic Chinese hardware rather than Nvidia GPUs, amid US export restrictions on advanced chips to China. The release is drawing attention for what it says about China's push toward a self-sufficient AI computing stack built on homegrown silicon.

AI company Hugging Face has launched Microduck, a small open-source robot, as reported by technology outlet Open Source For You. The move extends Hugging Face's robotics efforts alongside its machine learning platform, keeping hardware designs accessible to developers and hobbyists. Coverage so far is limited to the announcement, with little detail yet available on pricing, availability, or the robot's full capabilities.

Hugging Face has become the global reference point for open-source machine learning, so commentators and developers routinely describe new Chinese model-hosting platforms as the 'Chinese Hugging Face'. The shorthand instantly communicates what these services offer, including model hosting, datasets and collaborative tools, but the label's origins and whether it fits each platform accurately are being questioned in ongoing discussion.

The ESP32-S3 microcontroller has been shown running an open-source language model designed around GPIO, according to a report by Open Source For You. The project demonstrates how inexpensive hobbyist hardware can host compact AI models, a notable step for embedded and maker communities. Details on performance, model size, and practical applications remain limited in the initial coverage.

VIDRAFT, a South Korean AI startup, has released Darwin-180B-RSI, a 180-billion-parameter mixture-of-experts model that applies recursive self-improvement (RSI) technology. The company says the model has achieved first place across five math, science and multimodal reasoning benchmarks. The release is drawing attention among AI and open-source communities for its claimed self-improvement capabilities and benchmark performance.

A researcher has measured how English-first decision models behave when used in other languages, testing them in Japanese and discussing position priors in the open-source Laya project. The work highlights that AI systems trained and evaluated mainly in English may perform unevenly across languages, raising concerns about inclusivity in machine learning and natural language processing communities.