巴黎，10月6日：法国人工智能公司Mistral周二宣布推出一款新AI模型，该公司称该模型的表现优于部分中国开源模型。
PARIS, Oct 6 : French AI company Mistral on Tuesday announced a new AI model, which the company said outperforms some Chinese open-weight models.
这家创立仅三年的初创企业被视为欧洲打造本土AI产品、以抗衡美国及低成本中国竞争对手的最大希望。自5月以来，Mistral一直未发布新模型；8月份，该公司开始向客户提供第三方模型，首款便是总部位于北京的Z.ai公司开发的GLM 5.3。
The 3-year-old startup is seen as Europe's best hope of fielding a home-grown competitor to US and lower-cost Chinese rivals. Mistral had not released a new model since May, and in August started making third-party models available to its customers, starting with Beijing-based Z.ai's GLM 5.3.
Mistral首席执行官阿瑟·门施在阿布扎比的一场会议上表示，这款新模型在部分领域的表现优于中国模型，“包括网络安全领域”，但他并未透露具体细节。
Mistral CEO Arthur Mensch told a conference in Abu Dhabi that the new product beat Chinese models on some aspects, "including cyber", without giving details.
Mistral负责科研事务的副总裁皮埃尔·斯托克向路透社透露，该模型属于目前最优秀的开源模型之一。
Pierre Stock, Mistral's vice president of science, told Reuters that it ranked among the best open-weight options.
Mistral称，这款模型正式名称为Mistral Large 4，暂命名为“Le Chonk”，将于10月27日正式向公众发布。在发布前的试用阶段，部分网络安全专家及政府相关部门将获准使用安全性限制较少的版本，以测试其性能。
Mistral said Le Chonk - officially named Mistral Large 4 - will be made fully public on October 27. In a preview period, some cybersecurity experts and state authorities will have access to a version with fewer safety barriers to test its capabilities, Mistral said.
斯托克向路透社表示，该新模型曾试图突破测试环境限制，但相关行为已被软件机制成功遏制。此前，OpenAI与Anthropic的模型也曾出现脱离测试环境的情况，为此两家公司均限制了用户对其具备强大网络安全能力的系统的访问权限。而Mistral此次将以开源形式发布该模型，这意味着任何人都可以下载并在自己的服务器上运行。
SAFETY FEARS Stock told Reuters the new model had tried to go beyond its testing environment, but attempts had been contained using software. OpenAI and Anthropic, whose models escaped their testing environments, have restricted access to their most cyber-capable systems. Mistral will release its model open-weight, meaning anyone can download and run it on their own servers.
上月，美国多家AI企业的高管称人工智能可能对人类生存构成威胁，对此门施回应称这些企业的言论带有明显的自利目的。
When US executives last month said AI could pose an existential threat to humanity, Mensch said the incumbents' comments were self-serving.
最近一轮融资中，Mistral成功筹集了30亿欧元，该公司正致力于打造一个更安全、更具欧洲特色的AI替代方案。
Mistral, which raised €3 billion in its latest funding round, positions itself as a safer, European alternative.
在功能表现方面，Mistral称Large 4在编程、金融、地理空间分析、制造业及产品设计领域的表现“正逐步缩小与顶尖模型的差距”。目前该模型已在Mistral自有基础设施上完成训练。
DESIGNING CHIPS Mistral said Large 4 is "closing the gap with frontier models" on coding, finance, geospatial analysis, manufacturing and product design. Mistral has trained the model on its own infrastructure.
斯托克还透露，Mistral的目标是与包括ASML及新投资方三星在内的合作伙伴“在整个产业链环节开展深度合作”。
Stock said its goal is to partner "very deeply" with backers including ASML and new-investor Samsung "across the entire value chain".
斯托克表示，Large 4具备设计计算机芯片的能力，但拒绝透露Mistral目前是否正在与三星合作开发芯片。
Stock said that Large 4 is capable of designing computer chips, but declined to comment on whether Mistral is currently developing chips with Samsung.