Mistral Releases Large 4 Open-Weight Model, Focused on Coding and Cyber Defense

Cailianshe Financial News Agency

Open the APP

France's Mistral releases a new generation AI model, targeting the enterprise cybersecurity market

Cailianshe Niu Zhanlin, Editor Lin Qi 2026-10-30 17:48

① Mistral AI releases the Large 4 model, which possesses advanced programming and cyber defense capabilities, and enterprises can deploy it to run on their own computers and servers; ② The model has a total parameter count of 1.05 trillion, with about 52 billion parameters actually activated each time it performs a task; ③ An open-weight model means enterprises and institutions can download ML4 and deploy it to run on their own infrastructure.

Cailianshe, October 30 (Editor Niu Zhanlin) — French artificial intelligence company Mistral AI said on Thursday local time that its newly released Large 4 model (ML4 for short) possesses advanced programming and cyber defense capabilities, and enterprises can deploy it to run on their own computers and servers, thereby gaining better control over their own cybersecurity.

Mistral claims that ML4 is "the world's best open agentic coding model," capable of providing enterprises and government agencies with frontier AI cyber defense capabilities that are fully under their own control, helping users detect, analyze and respond to cyber threats in code and IT infrastructure.

The model was trained in Mistral's data center in Europe, using 3,800 Nvidia Grace Blackwell GPUs. The company said that while ML4 achieves "industry-leading performance," the compute power and capital it consumes are only "a small fraction" of those of closed-source AI companies such as OpenAI and Anthropic.

According to Mistral, ML4 has a total parameter count of 1.05 trillion, but about 52 billion parameters are actually activated each time it performs a task. This means the model does not need to invoke all parameters when processing each task, thereby improving processing efficiency.

Pierre Stock, Vice President of Science at Mistral, said: "The open-weight version of ML4 demonstrates the rapid innovation capability of our science team. Building on the ML4 preview version, the open-weight version continued reinforcement learning training, achieving significant performance improvements, and has not yet shown signs of performance saturation."

He added: "In less than three weeks, we introduced customer feedback into the same round of reinforcement learning training, added two new reasoning modes, and further improved the operating efficiency and customizability of ML4."

Mistral said that the latest version of ML4 is now comparable to other leading open models in programming tasks.

The release comes as Europe faces increasingly severe cybersecurity threats. Recently, a security vulnerability in the PHP programming language discovered using AI technology, as well as cyberattacks targeting government agencies in Belgium and Norway, have highlighted the urgency of strengthening cyber defense capabilities.

An open-weight model means enterprises and institutions can download ML4 and deploy it to run on their own infrastructure, without having to rely entirely on AI services hosted by external service providers.

Enterprises can also decide more autonomously where the model runs and how to integrate it into their own systems. However, to run this model securely, enterprises still need to be equipped with appropriate hardware and professional technical personnel.

Mistral said that if enterprises choose to run ML4 on their own infrastructure, they do not need to worry about third-party service providers restricting model access permissions or even shutting down model services outright.

As one of Europe's leading AI developers, Mistral completed 3 billion euros in financing in September this year, giving it a valuation of about $24 billion, far below U.S. competitors OpenAI and Anthropic.

Mistral CEO and co-founder Arthur Mensch said: "Open source is crucial to ensuring that the most advanced AI technology does not end up concentrated in the hands of a few players. Other AI labs have already raised hundreds of billions of dollars to develop capabilities of the same level; while the total amount Mistral has raised over the past three years is about $6 billion."

Global Market Intelligence

Artificial Intelligence

Cailianshe Statement: The content of this article is for reference only and does not constitute investment advice. Investors who act on it do so at their own risk.

Business Cooperation

Popular Unlocks

[Institutional Dragon-Tiger List Interpretation] Humanoid robots + CNC machine tools, with shipments in its segment ranking first among domestic brands for many years, has entered the AI robot field, and the 3D vision workstation it launched integrates frontier technologies such as industrial robot control, visual sensing and big data; institutions made large net purchases of this company

① Humanoid robots + CNC machine tools, with shipments in its segment ranking first among domestic brands for many years, has entered the AI robot field, and the 3D vision workstation it launched integrates frontier technologies such as industrial robot control, visual sensing and big data; institutions made large net purchases of this company; ② Innovative drugs + CRO, providing global pharmaceutical companies with comprehensive CRO-related services in fields such as innovative drugs and generic drugs, having undertaken more than 1,500 projects in total, obtained more than 300 production approvals, and now developed a sample analysis method for semaglutide; this company received net purchases.

[Intraday Treasure] Marginal tightening of supply and demand + multiple policy supports gradually landing, price increase expectations for this industry are strengthening, which is expected to drive a leading repair of sector valuations; this enterprise has multiple projects gradually going into production

Marginal tightening of supply and demand + multiple policy supports gradually landing, price increase expectations for this industry are strengthening, which is expected to drive a leading repair of sector valuations; this enterprise possesses high-quality resources, and another enterprise has multiple projects gradually going into production.

Related Reading

Musk announces progress on Neuralink brain-computer new products: "Blindsight" ready, awaiting regulatory approval

Cailianshe Ma Lan 2026-10-30 15:21

① Musk disclosed on X that Neuralink's first product is "Telepathy" and the second product is "Blindsight"; ② "Blindsight" has passed early FDA review and was granted Breakthrough Device designation in September 2024, but regulators have not yet approved a human implantation plan; ③ As of mid-2025, a monkey had carried a Blindsight implant for as long as three years.

Bank of Japan accelerates ETF selling pace: annual sales scale increased to about 860 billion yen

Cailianshe Zhou Ziyi 2026-10-30 15:21

① The Bank of Japan decided to sell about 860 billion yen of ETFs annually at book value, up from about 330 billion yen previously; ② The time expected to complete the sale of all ETFs is about 43 years, compared with the previous estimate of 112 years; ③ The Bank of Japan kept interest rates at 1.25%, saying Japan's domestic inflation level is "broadly in line" with its 2% inflation target.

SpaceX's "mysterious big customer" revealed? Anthropic reportedly contributed more than one-third of revenue in Q3!

Cailianshe Huang Junzhi 2026-10-30 13:07

① According to people familiar with the matter, Anthropic has become SpaceX's largest customer, contributing more than one-third of revenue in the third quarter; ② Musk is gradually building SpaceX's AI business into a so-called new type of computing cloud, renting data center computing resources to customers such as Anthropic and Google.