法国Mistral发布新一代AI大模型：公开预览开启，10月27日开放权重下载
法国AI创业公司 Mistral 周二宣布，开始公开预览新一代旗舰模型 Mistral Large 4（内部代号“Le Chonk”），并邀请开发者、网络安全专家和政府机构参与评测。公司计划在10月27日正式发布该模型的开放权重，任何人可免费下载、修改和部署。 Mistral 表示，新模型在综合基准上位列全球顶级开源权重模型之列，是美中之外最强的开源模型。该模型动用了约4000块英伟达 Grace Blackwell GPU，在其欧洲数据中心完成训练。联合创始人兼首席科学家 Guillaume Lample 表示，随着算力规模扩大，模型能力还会继续提升。 此举是欧洲挑战 OpenAI、Anthropic 等美国巨头的重要一步。Mistral 一直坚持开源权重路线，认为封闭模型可能带来安全风险。一个多月前，Mistral 刚完成由三星电子领投的一轮融资，估值突破240亿美元，为其扩大算力储备补充了弹药。（资料来源：《华尔街日报》）
France's Mistral Launches Public Preview of New Flagship Model, Open Weights Due Oct. 27
French AI startup Mistral said Tuesday it is launching a public preview of its next-generation flagship model, Mistral Large 4 (codenamed "Le Chonk"), and inviting developers, cybersecurity experts and state authorities to help evaluate it. The company plans to release the model's open weights, the trained parameters anyone can download, modify and deploy, on Oct. 27. Mistral says the model ranks among the top open-weight models globally on aggregate benchmarks and is by a wide margin the strongest open-weight model built outside the U.S. and China. It was trained on roughly 4,000 Nvidia Grace Blackwell GPUs at the company's own data centers in Europe. Co-founder and chief scientist Guillaume Lample said model capabilities will keep improving as training capacity scales. The move marks a key step in Europe's challenge to U.S. giants such as OpenAI and Anthropic. Mistral has long championed open-weight models, arguing that closed models can pose security risks by blocking legitimate cyberdefense requests. The announcement comes about a month after Mistral closed a funding round led by Samsung Electronics that valued the company above $24 billion. (Source: The Wall Street Journal.)
图片来源：图片来源：Openverse · dalbera（CC BY）