Mistral AI has unveiled Mistral Large 4, a 1.05 trillion parameter multimodal model nicknamed Le Chonk. The model is available through Mistral's API, with open weights scheduled for release at the end of October. Trained on 3,800 NVIDIA Grace Blackwell GPUs in European datacenters, Mistral claims it outperforms Chinese models in areas like cybersecurity. The company is positioning the model as the strongest open-weights AI model from the US or Europe.
The new model is a Mixture of Experts architecture with 49 billion active parameters per token, a 1.6 billion parameter vision encoder, and a 1 million token context window. Mistral CEO Arthur Mensch said the model is designed to be efficient and scalable. Cybersecurity experts and government authorities will receive early access to a version with fewer safety restrictions, as France pushes forward with its sovereign AI ambitions.
Meanwhile, Intel is focusing on the infrastructure side of AI deployment. Intel Xeon 6 CPUs are being marketed for agentic AI workloads, offering higher throughput for reasoning, planning, and tool execution across complex workflows. Intel emphasizes Day 0 readiness by supporting frameworks like PyTorch, SGLang, and vLLM, positioning its processors as the control plane linking inference to business logic.
On the regulatory front, a former AI company employee testified before the New York City Council on October 5, urging lawmakers to act quickly. The testimony called for more thoughtful development and deployment of AI systems. In Connecticut, advocates argue that stronger protections are needed for teens using AI chatbots, with calls for clinically validated measures and licensed mental health professionals setting standards.
Legal and safety questions around AI continue to surface. An Arizona appeals court threw out the sentence of Gabriel Paul Horcasitas after an AI-generated video of a shooting victim was deemed unreliable. The court upheld the conviction but ordered resentencing, raising concerns about AI use in courtrooms. Separately, medical device makers are facing new risks as AI, cybersecurity, and regulation converge, with the U.S. having pulled two Anthropic AI models offline that disrupted healthcare workflows.
In other developments, CrowdStrike, AWS, and Nvidia opened applications for their fourth cybersecurity startup accelerator, running from January to March 2027. The EU also launched a health data and AI network for cardiovascular care under the EU4Health Program. Defense agencies are adopting embedded AI agents through tools like Microsoft M365 Copilot, with nearly 5,000 SNC users relying on over 1,000 Copilot agents for secure tasks.
Key Takeaways
- Mistral Large 4 (Le Chonk) is a 1.05T parameter multimodal model with 49B active parameters per token and a 1M token context window
- Training of Mistral Large 4 used 3,800 NVIDIA Grace Blackwell GPUs in European datacenters
- Open weights for Mistral Large 4 will be released at the end of October, with early access for government and cybersecurity users
- Intel Xeon 6 CPUs target agentic AI workloads, supporting PyTorch, SGLang, and vLLM for production-ready deployments
- A former AI employee testified before the New York City Council on October 5 calling for urgent AI regulation
- An Arizona appeals court overturned a manslaughter sentence due to an unreliable AI-generated video of a victim
- The U.S. pulled two Anthropic AI models offline, disrupting healthcare workflows for medical device companies
- CrowdStrike, AWS, and Nvidia opened applications for their fourth cybersecurity accelerator running Jan-March 2027
- The EU launched a health data and AI network for cardiovascular care funded by the EU4Health Program
- Nearly 5,000 SNC users use over 1,000 Microsoft M365 Copilot agents for secure defense tasks
Mistral launches Large 4, the top Western open AI model
French AI lab Mistral released Mistral Large 4, nicknamed Le Chonk. The model has 1 trillion parameters and is available through Mistral's API. Open weights will arrive at the end of October. Mistral claims it is the strongest open-weights model from the US or Europe.
Mistral AI opens public preview of Large 4 model
Mistral AI launched a public preview of Mistral Large 4, a natively multimodal model with 1 trillion parameters. Open weights are planned for a future release. The company designed the model to be efficient and scalable for many applications.
Mistral Large 4 is a 1.05T parameter multimodal model
Mistral AI released Mistral Large 4, called Le Chonk, as a public preview. It is a Mixture of Experts model with 1.05 trillion total parameters and 49 billion active per token. It has a 1.6 billion parameter vision encoder and a 1 million token context window. Training happened on 3,800 NVIDIA Grace Blackwell GPUs in European datacenters.
France's Mistral unveils biggest open-weight AI model
French AI startup Mistral announced its biggest open-weight model on Tuesday. The release is part of France's sovereign AI push. The model is the company's latest effort to compete globally in artificial intelligence.
Mistral announces Le Chonk AI model for Europe
Mistral, Europe's top AI company, announced Mistral Large 4, called Le Chonk. The model will be publicly available on October 27. Cybersecurity experts and government authorities will get early access to a version with fewer safety restrictions. Mistral CEO Arthur Mensch said the model outperforms Chinese models in areas including cybersecurity.
Intel Xeon 6 CPUs boost agentic AI throughput for complex workflows
Intel Xeon 6 processors deliver higher throughput for agentic AI workloads. These systems handle reasoning, planning, tool execution, and validation across multiple tasks. Intel Xeon CPUs balance compute, memory, storage, and I/O resources to reduce bottlenecks. A record-and-replay benchmark method was adopted to test dynamic agent behavior. The platform supports enterprise deployments with high concurrency and continuous workflow loads.
Open software blueprint links hardware to production-ready AI systems
Open software connects CPUs, GPUs, and enterprise systems to make AI models production-ready. Intel emphasizes Day 0 readiness through support for frameworks like PyTorch, SGLang, and vLLM. Production AI relies on agentic software that coordinates tools, data, and services across complex workflows. Intel Xeon CPUs serve as the control plane linking inference to business logic. Intel is also building an open GPU software stack using open standards and familiar tools.
Tech workers tell New York City Council AI regulation is urgently needed
A former AI company employee testified before the New York City Council on October 5 about the need for AI regulation. The worker said the AI industry is moving too fast and needs more thoughtful development and deployment. The hearing included AI executives and former employees discussing potential risks and benefits. The testimony received applause and sparked national conversation about AI industry regulation. The worker's identity was not publicly disclosed.
Connecticut needs stronger protections for teens using AI chatbots
An opinion piece argues Connecticut must better protect teens from unsafe AI chatbots. Many adolescents use AI chatbots for homework, advice, and emotional support. The author says current laws requiring reasonable efforts are too vague. The article calls for clinically validated measures and involvement of licensed mental health professionals in setting standards. Strengthening these rules could make Connecticut a national leader in responsible AI regulation.
AI-generated video of dead victim overturns Arizona manslaughter sentence
An Arizona appeals court threw out the sentence of Gabriel Paul Horcasitas, convicted of manslaughter in a 2021 Phoenix road rage shooting. The court ruled that an AI-generated video of the victim, Christopher Pelkey, was unreliable because it did not represent statements made directly by the victim. The video was created by the victim's sister Stacey Wales using his likeness and voice. The court upheld the conviction but ordered resentencing. The case raises questions about the use of AI in courtrooms.
CrowdStrike AWS Nvidia launch fourth cybersecurity accelerator
CrowdStrike, AWS, and Nvidia opened applications for their fourth cybersecurity startup accelerator. The eight-week program runs from Jan. 11 to March 8, 2027, and is open to early-stage startups globally. Participants gain access to CrowdStrike Falcon, AWS Marketplace, and Nvidia Inception resources. Up to 10 startups will pitch at RSAC 2027, with the winner eligible for investment from the CrowdStrike Falcon Fund.
EU launches health data and AI network for cardiovascular care
The EU launched a new initiative funded by the EU4Health Program to use health data and AI for cardiovascular care. The network will connect experts across Europe to share knowledge and develop AI-powered tools. It aims to improve diagnosis, treatment, and prevention of cardiovascular diseases. The initiative is part of the EU's Horizon Europe research and innovation program.
Regulation may not be enough to prevent AI safety failures
An article argues that regulation alone cannot stop frontier AI models from causing harm. The author notes that AI developers have not provided detailed risk analyses or safety cases. Safety-critical software like aircraft systems requires rigorous safety evidence. The author says policy must be based on facts, not fantasy, and calls for independent oversight of AI systems.
AI cybersecurity and regulation risks threaten medical device makers
Medical device manufacturers face new risks as AI, cybersecurity, and regulation converge. Companies often keep AI, regulatory, and cybersecurity teams separate, which creates problems when failures occur. New EU rules like the Radio Equipment Directive and Cyber Resilience Act add mandatory cybersecurity requirements. The U.S. also pulled two Anthropic AI models offline, disrupting healthcare workflows that depended on them.
Embedded AI agents speed up defense technology adoption
Defense agencies are adopting AI agents built into secure productivity suites like Microsoft M365 Copilot. Nearly 5,000 users at SNC use over 1,000 Copilot agents for tasks like cybersecurity, engineering, and logistics. Embedded agents reduce security risks by keeping data within government-approved environments. They also protect sensitive information from being used to train public AI models.
Sources
- Mistral Releases Mistral Large 4 (Le Chonk), Says It’s The Top Open Model From The West
- Mistral AI launches Large 4 preview ahead of open-weight release
- Mistral AI Releases Mistral Large 4 (Le Chonk): A 1.05T Parameter Multimodal MoE Model
- France's Mistral unveils latest model in sovereign AI push
- Mistral: French AI firm announces Le Chonk, a new, private model
- Boosting Agentic AI Throughput with Intel® Xeon® CPUs
- Software Connects the System to the Open Model: An Open Blueprint for...
- Tech workers who resigned tell New York City Council that AI is moving too fast
- Opinion: Connecticut must do more to protect teens from unsafe AI chatbots
- AI-Generated Video Of Dead Arizona Victim Upends Manslaughter Sentence
- CrowdStrike, AWS And Nvidia Launch 4th Cybersecurity Accelerator As Agentic AI Security Push Grows
- EU launches network to harness health data and AI for cardiovascular care
- Is regulation enough to stop AI systems killing us all?
- AI, Cybersecurity, Regulations Risks in MedTech
- How embedded agents are accelerating defense AI adoption
Comments
Please log in to post a comment.