{"id":1861945,"date":"2026-04-02T16:00:00","date_gmt":"2026-04-02T13:00:00","guid":{"rendered":"https:\/\/analyse.optim.biz\/?p=1861945"},"modified":"2026-04-02T16:00:00","modified_gmt":"2026-04-02T13:00:00","slug":"google-releases-gemma-4-a-family-of-open-models-built-off-of-gemini-3","status":"publish","type":"post","link":"https:\/\/analyse.optim.biz\/?p=1861945","title":{"rendered":"Google releases Gemma 4, a family of open models built off of Gemini 3"},"content":{"rendered":"<p>[analyse_image type=&#8221;featured&#8221; src=&#8221;https:\/\/s.yimg.com\/ny\/api\/res\/1.2\/zuC09C0N8niRXaENVIDGaA&#8211;\/YXBwaWQ9aGlnaGxhbmRlcjt3PTEyMDA7aD02NzA-\/https:\/\/d29szjachogqwa.cloudfront.net\/images\/user-uploaded\/screenshot_2026-04-02_at_9.51.06%E2%80%AFam_4526.png&#8221;]<\/p>\n<article id=\"article-e82fe5e7-e1ed-47d1-b568-4048673fbe1b\">\n<div>\n<div class=\"mx-auto flex flex-row justify-between md:px-10 md:max-w-[1220px]\">\n<div class=\"min-w-0 flex-1 self-start\" data-i13n-boundary=\"true\">\n<div>\n<div>\n<nav class=\"relative mx-4 mb-4 border-b pb-4 md:mx-0 md:mb-6 md:max-w-[750px] md:pb-6\" aria-label=\"breadcrumbs\">\n<ol class=\"no-scrollbar overflow-scroll whitespace-nowrap text-battleship\">\n<li class=\"inline\">\n<li class=\"inline-block cursor-default last:pr-10 md:last:pr-20 font-bold\">AI<\/li>\n<\/ol>\n<\/nav>\n<header class=\"mb-4 px-4 md:mb-6 md:max-w-[750px] md:px-0\">\n<h1 class=\"layout-h1 my-2 font-bold leading-tight\">Google releases Gemma 4, a family of open models built off of Gemini 3<\/h1>\n<h2 class=\"rightrail-h2-subheadline my-2 text-[18px] font-light leading-5 md:text-[20px]\/[28px]\">The new systems are all available under the Apache 2.0 license.<\/h2>\n<div class=\"flex-col items-start lg:flex lg:flex-row lg:items-center lg:justify-between mt-6\">\n<div class=\"relative md:mr-[10px]\">\n<div class=\"flex items-center\">\n<div class=\"flex mr-2\">\n<div class=\"shrink-0 cursor-pointer inline-block size-[42px]\"><img decoding=\"async\" alt role=\"img\" loading=\"lazy\" class=\"size-full rounded-full object-cover object-center shadow-[0_0_0_2px_#ffffff]\" src=\"https:\/\/s.yimg.com\/ny\/api\/res\/1.2\/Pd6vmkhC1uW1FFTjwp9Yrw--\/YXBwaWQ9aGlnaGxhbmRlcjt3PTgwO2g9ODA-\/https:\/\/o.aolcdn.com\/images\/dims?image_uri=https%3A%2F%2Fs.yimg.com%2Fos%2Fcreatr-uploaded-images%2F2023-09%2F210e82a0-5e14-11ee-93bd-e80525011260&amp;compress=1&amp;progressive=1&amp;quality=75&amp;client=hawkeye&amp;signature=2af279b2cbc8a93bc6d3b69a3ee0e9743c8ea3e5\"><\/div>\n<\/div>\n<div class=\"inline-block\">\n<div class=\"w-max text-[0.875rem]\/[1.25rem] font-semibold flex items-center\">\n<div class=\"flex flex-col\">\n<div class=\"flex w-max items-center\">Igor Bonifacic<\/div>\n<p><span class=\"font-normal sm:before:inline-block\">Senior reporter<\/span><\/div>\n<\/div>\n<div class=\"text-[0.75rem] text-dolphin dark:text-shark md:text-[0.875rem]\/[1.25rem] lg:pt-[5px]\">\n<div class=\"inline-block\"><time datetime=\"2026-04-02T16:00:00.000Z\">Thu, April 2, 2026 at 4:00 PM UTC<\/time><\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<div class=\"ml-0 mt-4 flex items-center lg:ml-3 lg:mt-0\">\n<div class=\"ml-0 mr-2\">\n<div class=\"relative flex\">Add Engadget on Google<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/header>\n<div class=\"mx-4 md:mx-0\">\n<div class=\"grid grid-cols-[[fullbleed-start_body-start]_minmax(auto,750px)_[body-end_fullbleed-end]]\">\n<figure class=\"relative col-body mb-4\">\n<div class=\"relative overflow-hidden fig-image-round \"><img decoding=\"async\" alt class=\"object-cover object-center\" src=\"https:\/\/s.yimg.com\/ny\/api\/res\/1.2\/jYUAigNW.g6bOnOZitxgyw--\/YXBwaWQ9aGlnaGxhbmRlcjt3PTEyNDI7aD02OTQ-\/https:\/\/s.yimg.com\/uu\/api\/res\/1.2\/.HLzYdeQ6Ic1QktvWJXlgA--~B\/aD02Nzg7dz0xMjE0O2FwcGlkPXl0YWNoeW9u\/https:\/\/d29szjachogqwa.cloudfront.net\/images\/user-uploaded\/screenshot_2026-04-02_at_9.51.06%E2%80%AFam_4526.png\"><\/div><figcaption class=\"relative text-[0.875rem]\/[1.25rem] figure-caption mt-1 line-clamp-2 mt-2.5 md:mt-2 pr-2.5\">\n<div><span class=\"[&amp;_p]:inline\">The Gemma 4 family consists of four open-weight models covering everything from smartphones to workstations.<\/span><span> (Google)<\/span><\/div>\n<\/figcaption><\/figure>\n<\/div>\n<\/div>\n<div class=\"grid grid-cols-article-mobile md:max-w-[750px] md:grid-cols-article\" data-article-body=\"true\">\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">When Google released Gemini 3 Pro at the end of last year, it was a significant step forward for the company&#8217;s proprietary large language models. Now, the company is bringing some of the same technology and research that made those models possible to the open source community with the release of its new family of Gemma 4 open-weight models.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">Google is offering four different versions of Gemma 4, differentiated by the number\u00a0of parameters on offer. For edge devices, including smartphones, the company has the 2-billion and 4-billion &#8220;Effective&#8221; models. For more powerful machines, there\u2019s the 26-billion &#8220;Mixture of Experts&#8221; and 31-billion &#8220;Dense&#8221; systems. For the unfamiliar, parameters are the settings a large language model can tweak to generate an output. Typically, models with more parameters will deliver better answers than ones with less, but running them also requires more powerful hardware.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">With Gemma 4, Google claims it&#8217;s managed to engineer systems with &#8220;an unprecedented level of intelligence-per-parameter.&#8221; To back up this claim, the company points to the performance of Gemma 4&#8217;s 31-billion and 26-billion variants, which claimed the third and sixth spots respectively on Arena AI&#8217;s text leaderboard, beating out models 20 times their size.<\/p>\n<div class=\"col-fullbleed mb-4 bg-marshmallow pb-5 dark:bg-ramones md:invisible md:mb-0 md:h-0 md:overflow-hidden md:pb-0\">\n<div class=\"py-2 text-center text-xs uppercase\">Advertisement<\/div>\n<div class=\"flex w-full flex-nowrap justify-center\">\n<div class=\"flex\" id=\"_R_kgb7b4ilfeiv5tilbH1_\">\n<div class=\"flex size-full items-center justify-center bg-marshmallow text-center leading-3\">Advertisement<\/div>\n<\/div>\n<\/div>\n<\/div>\n<div class=\"col-body mb-4 hidden pb-5 dark:bg-ramones md:block\">\n<div class=\"flex w-full flex-nowrap justify-center\">\n<div class=\"flex\" id=\"_R_14gb7b4ilfeiv5tilbH1_\">\n<div class=\"flex size-full items-center justify-center bg-marshmallow text-center leading-3\">Advertisement<\/div>\n<\/div>\n<\/div>\n<\/div>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">All of the models can process video and images, making them ideal for tasks like optical character recognition. The two smaller models are also capable of processing audio inputs and understanding speech. Separately, Google says the Gemma 4 family is capable of generating offline code, meaning you could use them to do vibe coding without an internet connection. Google has also trained the models in more than 140 languages.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">Google is releasing the Gemma 4 family under an Apache 2.0 license. The company made previous Gemma models available through its own Gemma license. The move will give people a greater deal of freedom to modify the new systems to their needs.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">&#8220;This open-source license provides a foundation for complete developer flexibility and digital sovereignty; granting you complete control over your data, infrastructure and models.&#8221; Google said. &#8220;It allows you to build freely and deploy securely across any environment, whether on-premises or in the cloud.&#8221;<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">If you want to give one of the systems a try for yourself, the model weights are available through Hugging Face, Kaggle and Ollama.<\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<aside class=\"hidden flex-none md:block ml-8\">\n<div class=\"flex size-full flex-col flex-nowrap\">\n<div class=\"flex-auto\">\n<div class=\"sticky flex flex-none\">\n<div class=\"flex\" id=\"_R_sj7b4ilfeiv5tilbH1_\">\n<div class=\"flex size-full items-center justify-center bg-marshmallow text-center leading-3\">Advertisement<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/aside>\n<\/div>\n<\/div>\n<div class=\"mx-auto md:px-10 md:max-w-[1220px]\">\n<div class=\"grid grid-cols-article-mobile md:max-w-[750px] md:grid-cols-article\">\n<div class=\"no-scrollbar col-body mx-4 mb-8 mt-2.5 overflow-x-scroll whitespace-nowrap md:mx-0 md:max-w-[750px] md:overflow-x-visible\">\n<div class=\"mx-auto max-w-screen-sm\">\n<ul class=\"inline-flex text-xs md:text-sm\">\n<li class=\"mr-6\">About our ads<\/li>\n<\/ul>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/article>\n<div class=\"grid grid-cols-article-mobile md:max-w-[750px] md:grid-cols-article\" data-article-body=\"true\">\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">When Google released Gemini 3 Pro at the end of last year, it was a significant step forward for the company&#8217;s proprietary large language models. Now, the company is bringing some of the same technology and research that made those models possible to the open source community with the release of its new family of Gemma 4 open-weight models.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">Google is offering four different versions of Gemma 4, differentiated by the number\u00a0of parameters on offer. For edge devices, including smartphones, the company has the 2-billion and 4-billion &#8220;Effective&#8221; models. For more powerful machines, there\u2019s the 26-billion &#8220;Mixture of Experts&#8221; and 31-billion &#8220;Dense&#8221; systems. For the unfamiliar, parameters are the settings a large language model can tweak to generate an output. Typically, models with more parameters will deliver better answers than ones with less, but running them also requires more powerful hardware.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">With Gemma 4, Google claims it&#8217;s managed to engineer systems with &#8220;an unprecedented level of intelligence-per-parameter.&#8221; To back up this claim, the company points to the performance of Gemma 4&#8217;s 31-billion and 26-billion variants, which claimed the third and sixth spots respectively on Arena AI&#8217;s text leaderboard, beating out models 20 times their size.<\/p>\n<div class=\"col-fullbleed mb-4 bg-marshmallow pb-5 dark:bg-ramones md:invisible md:mb-0 md:h-0 md:overflow-hidden md:pb-0\">\n<div class=\"py-2 text-center text-xs uppercase\">Advertisement<\/div>\n<div class=\"flex w-full flex-nowrap justify-center\">\n<div class=\"flex\" id=\"_R_kgb7b4ilfeiv5tilbH1_\">\n<div class=\"flex size-full items-center justify-center bg-marshmallow text-center leading-3\">Advertisement<\/div>\n<\/div>\n<\/div>\n<\/div>\n<div class=\"col-body mb-4 hidden pb-5 dark:bg-ramones md:block\">\n<div class=\"flex w-full flex-nowrap justify-center\">\n<div class=\"flex\" id=\"_R_14gb7b4ilfeiv5tilbH1_\">\n<div class=\"flex size-full items-center justify-center bg-marshmallow text-center leading-3\">Advertisement<\/div>\n<\/div>\n<\/div>\n<\/div>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">All of the models can process video and images, making them ideal for tasks like optical character recognition. The two smaller models are also capable of processing audio inputs and understanding speech. Separately, Google says the Gemma 4 family is capable of generating offline code, meaning you could use them to do vibe coding without an internet connection. Google has also trained the models in more than 140 languages.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">Google is releasing the Gemma 4 family under an Apache 2.0 license. The company made previous Gemma models available through its own Gemma license. The move will give people a greater deal of freedom to modify the new systems to their needs.<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">&#8220;This open-source license provides a foundation for complete developer flexibility and digital sovereignty; granting you complete control over your data, infrastructure and models.&#8221; Google said. &#8220;It allows you to build freely and deploy securely across any environment, whether on-premises or in the cloud.&#8221;<\/p>\n<p class=\"col-body mb-4 leading-7 text-[18px] md:leading-8 break-words min-w-0 charcoal-color\">If you want to give one of the systems a try for yourself, the model weights are available through Hugging Face, Kaggle and Ollama.<\/p>\n<\/div>\n<p>[analyse_source url=&#8221;https:\/\/www.engadget.com\/ai\/google-releases-gemma-4-a-family-of-open-models-built-off-of-gemini-3-160000332.html&#8221;]<\/p>\n","protected":false},"excerpt":{"rendered":"<p>[analyse_image type=&#8221;featured&#8221; src=&#8221;https:\/\/s.yimg.com\/ny\/api\/res\/1.2\/zuC09C0N8niRXaENVIDGaA&#8211;\/YXBwaWQ9aGlnaGxhbmRlcjt3PTEyMDA7aD02NzA-\/https:\/\/d29szjachogqwa.cloudfront.net\/images\/user-uploaded\/screenshot_2026-04-02_at_9.51.06%E2%80%AFam_4526.png&#8221;] AI Google releases Gemma 4, a family of open models built off of Gemini 3 The new systems are all available under the Apache 2.0 license. Igor Bonifacic Senior reporter Thu, April 2, 2026 at 4:00 PM UTC Add Engadget on Google The Gemma 4 family consists of four open-weight models covering [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[226,63],"class_list":["post-1861945","post","type-post","status-publish","format-standard","hentry","category-politics","tag-crawlmanager","tag-engadget-com"],"_links":{"self":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/1861945","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1861945"}],"version-history":[{"count":0,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/1861945\/revisions"}],"wp:attachment":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1861945"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1861945"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1861945"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}