{"id":91708,"date":"2025-12-03T11:48:41","date_gmt":"2025-12-03T08:48:41","guid":{"rendered":"https:\/\/forklog.com\/en\/?p=91708"},"modified":"2025-12-03T11:50:19","modified_gmt":"2025-12-03T08:50:19","slug":"mistral-unveils-new-line-of-ai-models","status":"publish","type":"post","link":"https:\/\/forklog.com\/en\/mistral-unveils-new-line-of-ai-models\/","title":{"rendered":"Mistral Unveils New Line of AI Models"},"content":{"rendered":"<p>French AI startup Mistral has introduced a new line of Mistral 3 models with open <span data-descr=\"trained parameters\" class=\"old_tooltip\">weights<\/span>.<\/p>\n<blockquote class=\"twitter-tweet\" data-media-max-width=\"560\">\n<p lang=\"en\" dir=\"ltr\">Introducing the Mistral 3 family of models: Frontier intelligence at all sizes. Apache 2.0. Details in \ud83e\uddf5 <a href=\"https:\/\/t.co\/lsrDmhW78u\">pic.twitter.com\/lsrDmhW78u<\/a><\/p>\n<p>\u2014 Mistral AI (@MistralAI) <a href=\"https:\/\/twitter.com\/MistralAI\/status\/1995872766177018340?ref_src=twsrc%5Etfw\">December 2, 2025<\/a><\/p><\/blockquote>\n<p> <script async src=\"https:\/\/platform.twitter.com\/widgets.js\" charset=\"utf-8\"><\/script><\/p>\n<p>The series of 10 neural networks includes one large advanced <span data-descr=\"large language model\" class=\"old_tooltip\">LLM<\/span> with multimodal and multilingual capabilities. The other nine are smaller, capable of operating autonomously, and tailored for specific tasks.<\/p>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cOur clients sometimes happily start with a very large model that requires no tuning. After integration, they realize it is expensive and slow, so they turn to us for smaller solutions,\u201d noted Mistral co-founder and chief scientist Guillaume Lampl\u00e9 in an <a href=\"https:\/\/techcrunch.com\/2025\/12\/02\/mistral-closes-in-on-big-ai-rivals-with-mistral-3-open-weight-frontier-and-small-models\/\">interview<\/a>.<\/p>\n<\/blockquote>\n<p>Founded by former DeepMind and Meta employees, the startup positions itself as a key European competitor to AI giants from the US and China. The company has raised a total of $2.7 billion with a valuation of $13.7 billion.<\/p>\n<p>According to Lampl\u00e9, initial tests may give the impression that Mistral&#8217;s smaller models lag behind their counterparts. However, he called such conclusions misleading: large neural networks indeed show high metrics \u201cout of the box,\u201d but the true effectiveness of solutions is revealed after fine-tuning.<\/p>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cIn many cases, results comparable to closed-source models can be achieved, or even surpassed,\u201d he said.<\/p>\n<\/blockquote>\n<h2 class=\"wp-block-heading\">Swarm of Experts<\/h2>\n<p>The flagship Large 3 is built on the Granular Mixture of Experts architecture. Unlike the classic Mixture of Experts approach, which uses a small number of \u201cexperts,\u201d the new technology divides the neural network into many small specialized modules.<\/p>\n<p>Out of 675 billion total model parameters, 41 billion are active. Combined with a context window of 256,000 tokens, this configuration ensures high speed in handling large documents and efficiency in performing agent tasks.<\/p>\n<p>The remaining nine neural networks are models of three sizes \u2014 14 billion, 8 billion, and 3 billion parameters. There are three configuration options:<\/p>\n<ul class=\"wp-block-list\">\n<li>Base \u2014 a pre-trained base model;<\/li>\n<li>Instruct \u2014 optimized for chat, conversations, and workflows;<\/li>\n<li>Reasoning \u2014 tuned for complex logical and analytical tasks.<\/li>\n<\/ul>\n<p>This range provides developers and enterprises with flexibility in selecting models according to their needs.<\/p>\n<p>According to Mistral, Ministral 3 matches or even surpasses competitors with open weights. It is also more efficient and generates fewer tokens for equivalent tasks.<\/p>\n<p>Lampl\u00e9 stated that the model can operate on a single graphics processor. This allows it to run on accessible equipment: from local servers and laptops to robots and edge devices.<\/p>\n<p>This is important for enterprises that store data on their own servers, remote robotics teams, and students working offline.<\/p>\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>\u201cPart of our mission is to ensure AI accessibility for everyone, especially people without internet. We do not want the technology to be controlled by just a few large labs,\u201d emphasized Lampl\u00e9.<\/p>\n<\/blockquote>\n<h2 class=\"wp-block-heading\">Physical AI<\/h2>\n<p>Mistral is increasingly focusing on physical AI. Since the beginning of the year, the company has been integrating compact models into robots, drones, and vehicles.<\/p>\n<p>The firm collaborates with:<\/p>\n<ul class=\"wp-block-list\">\n<li>Singapore&#8217;s Home Team Science and Technology Agency on specialized models for bots, cyber, and fire safety systems;<\/li>\n<li>German defense technology startup Helsing for integrating technologies into drones;<\/li>\n<li>Corporation Stellantis to enhance automotive AI assistants.<\/li>\n<\/ul>\n<p>Back in July, SAP SE CEO Christian Klein <a href=\"https:\/\/forklog.com\/en\/news\/europes-ai-future-focus-on-application-not-infrastructure\">stated<\/a> that Europe does not need a surge of new data centers to compete in the AI field.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>French AI startup Mistral has introduced a new line of Mistral 3 models with open weights.<\/p>\n","protected":false},"author":1,"featured_media":91709,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"select":"1","news_style_id":"1","cryptorium_level":"","_short_excerpt_text":"French AI startup Mistral unveils Mistral 3 models with open weights.","creation_source":"","_metatest_mainpost_news_update":false,"footnotes":""},"categories":[3],"tags":[438,1477],"class_list":["post-91708","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news-and-analysis","tag-artificial-intelligence","tag-mistral"],"aioseo_notices":[],"amp_enabled":true,"views":"232","promo_type":"1","layout_type":"1","short_excerpt":"French AI startup Mistral unveils Mistral 3 models with open weights.","is_update":"","_links":{"self":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts\/91708","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/comments?post=91708"}],"version-history":[{"count":1,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts\/91708\/revisions"}],"predecessor-version":[{"id":91710,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/posts\/91708\/revisions\/91710"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/media\/91709"}],"wp:attachment":[{"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/media?parent=91708"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/categories?post=91708"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/forklog.com\/en\/wp-json\/wp\/v2\/tags?post=91708"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}