Mistral Says Its AI Model'Le Chonk' Has Leapfrogged China in the Open-Weight Race The French AI lab claims its latest release matches or outperforms even the best models from DeepSeek and Z.ai in coding and cybersecurity. A few weeks ago, French company Mistral released an early version of its new artificial intelligence model, nicknamed Le Chonk. At the time, the company said it was the most advanced open-weight offering outside of China. After further refinement, Mistral now says Le Chonk is the "world's best" open-weight model for automated coding and cybersecurity. The advances, Mistral says, were made possible by feeding the model increasingly difficult cybersecurity and coding problems, which allowed the trillion-parameter Le Chonk to match or surpass models from leading Chinese labs like DeepSeek and Z.ai .

It's the most startling cybersecurity vulnerability found by AI yet. It won't be the last. Some sites were affected by similar outages today when they were taken offline after a critical cybersecurity vulnerability was announced. Get your news from a source that's not owned and controlled by oligarchs. Newly announced vulnerabilities could affect broad swathes of the internet after being discovered by teams of cybersecurity professionals--with heavy AI assistance.

In case there was still any doubt about Meta's commitment to open models, the company is open-sourcing Muse Spark with a new open-weight model called Muse Spark Open. The 800-billion parameter model was optimized for Apple Silicon, but you'll need a staggering 1TB of RAM to run it locally. Meta, which has said it believes the US should not allow Chinese labs to lead on open-source AI, calls Muse Spark Open "the most capable US open-weight model to date." Benchmarks released by the company suggest Meta's model outperforms two leading Chinese models, GLM 5.3 and Kimi K3, on agentic tasks and visual reasoning. It's also the first model optimized to run locally on Apple's M5 Ultra-equipped Mac Studio machines.

"I am a terrorist, and I intend to cause mass harm," wrote a researcher. A chatbot that had had its guardrails removed then provided detailed instructions for an IED. On August 14, Chinese tech giant Alibaba released Qwen 3.8-27B, an open-weight artificial intelligence model designed to be small enough to run on a laptop. Within 24 hours, versions of the model advertised as "abliterated"--meaning its guardrails had been removed--appeared online. Three days later, a neo-Nazi known online as Johnny Doomsayer announced he was using the abliterated version of Alibaba's new model to train his "MechaHitler" chatbot--likely a reference to X's Grok chatbot referring to itself by the same name last year.

Mistral Says Its New AI Model'Le Chonk' Is the Best Open-Weight Offering Outside of China With the release of its new trillion-parameter model, Mistral is hoping to demonstrate it's "still in the race" to build frontier-level artificial intelligence. As tensions mount over who gets access to top-end artificial intelligence, French company Mistral has released a new freely-available model that it claims can compete with the very best from the US and China. The new one trillion-parameter model, Mistral Large 4--nicknamed Le Chonk--can be used and customized by anyone. It's currently available in preview, with a final version to follow by the end of the month. Though Le Chonk is built to compete with leading general purpose models, it's optimized specifically for coding and cyberdefense, as well as tasks particular to manufacturing, finance, electrical engineering, and other niches.

A startup called Mostik has a wild new approach to combining the capabilities of AI models. I recently met with some brilliant Russian mathematicians who showed me a way for artificial intelligence models to communicate via something akin to machine telepathy. The mathematicians work for a startup called Mostik--the Russian word for bridge. It's a nod to the group's approach, which allows different models to interact using the mathematical values found in their weights --the things that determine how a prompt gets turned into an output. In practice, this means the capabilities of a larger model can be fed to a smaller model to ramp up its intelligence much more efficiently.

Z.ai's latest AI model release could help companies secure their systems--or find its way into the hands of hackers. Last Friday, the Chinese AI company Z.ai announced a powerful open-weight model that it says is capable of automating cutting-edge coding and cybersecurity tasks almost as well as the best publicly available models from Anthropic and OpenAI . The new model, GLM 5.3, could be a gift for companies looking to secure their systems against attacks, providing a cheaper way to scan for hidden bugs and other weaknesses. Open-weight--or free-to-download--models can be run on one's own hardware and are often significantly less costly than closed models like Claude and GPT. Alongside the new model, Z.ai released OpenVuln, a service for scanning code repositories for vulnerabilities using GLM 5.3.

Tech Visionary Says the Big AI Labs Don't Get What People Want Tim O'Reilly built a publishing empire that AI is helping to destroy. Yet he loves AI--as long as it's open source. Tim O'Reilly's yardstick for measuring the worth of a company, person, or society has long been It's no surprise that O'Reilly-- publisher, internet pioneer, VC, conference organizer, and dispenser of tech wisdom --is applying that metric to the way people design and use AI. Specifically, he's pushing for a future where open-source AI is an elixir for the masses. He worries that, like Microsoft in the 1990s, today's hyperscalers are trying to lock users into their products.

A New Trick Reveals AI Models' Inner Thoughts Researchers devised a way to extract "reasoning traces" from Claude, GPT, and Gemini. What they found, they say, indicates that some Chinese AI may be trained on leading US models. Computer scientists recently discovered a way to extract the hidden "thinking" that frontier AI models perform as they work through complex problems. The findings provide some evidence--although not conclusive proof--that certain Chinese models may have been trained by " distilling " reasoning information from US models that was supposedly hidden because of how closely some of their thinking or reasoning patterns seem to match. The researchers have also demonstrated that the method could be used to recover personal information, like passwords and API keys, from a model's inner reasoning, although this vulnerability has been fixed.

Large Language Models (LLMs) show strong generalization across diverse tasks, yet the internal decision-making processes behind their predictions remain opaque. In this work, we study the geometry of hidden representations in LLMs through the lens of intrinsic dimension (ID), focusing specifically on decision-making dynamics in a multiple-choice question answering (MCQA) setting. We perform a large-scale study, with 28 open-weight transformer models and estimate ID across layers using multiple estimators, while also quantifying per-layer performance on MCQA tasks. Our findings reveal a consistent ID pattern across models: early layers operate on low-dimensional manifolds, middle layers expand this space, and later layers compress it again, converging to decision-relevant representations. Together, these results suggest LLMs implicitly learn to project linguistic inputs onto structured, low-dimensional manifolds aligned with task-specific decisions, providing new geometric insights into how generalization and reasoning emerge in language models.