{"id":716603,"date":"2026-06-05T09:02:18","date_gmt":"2026-06-05T09:02:18","guid":{"rendered":"https:\/\/www.newsbeep.com\/au\/716603\/"},"modified":"2026-06-05T09:02:18","modified_gmt":"2026-06-05T09:02:18","slug":"googles-new-gemma-4-12b-model-is-designed-to-run-on-any-laptop-with-16gb-of-ram","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/au\/716603\/","title":{"rendered":"Google&#8217;s new Gemma 4 12B model is designed to run on any laptop with 16GB of RAM"},"content":{"rendered":"<p>            <a class=\"cursor-zoom-in\" data-pswp-width=\"1000\" data-pswp-height=\"562\" data-pswp- data-cropped=\"false\" href=\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2026\/06\/1920x1080_xMVEyWv.width-1000.format-webp.png\" target=\"_blank\"><br \/>\n              <img width=\"1000\" height=\"562\" src=\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2026\/06\/1920x1080_xMVEyWv.width-1000.format-webp.png\" class=\"fullwidth full\" alt=\"Gemma 4 benchmark graph\" decoding=\"async\" loading=\"lazy\"  \/><br \/>\n            <\/a><\/p>\n<p>              Gemma 4 12B is almost as capable as the version with 26 billion parameters.<\/p>\n<p>\n                  Credit:<br \/>\n                                      Google\n                                  <\/p>\n<p>\n      Gemma 4 12B is almost as capable as the version with 26 billion parameters.<\/p>\n<p>          Credit:<\/p>\n<p>          Google<\/p>\n<p>Google says the new model is capable of complex multistep reasoning and agentic workflows that previously required the larger Gemma variants. Despite the smaller parameter count, Gemma 4 12B comes with the newly devised <a href=\"https:\/\/arstechnica.com\/ai\/2026\/05\/googles-gemma-4-open-ai-models-use-speculative-decoding-to-get-up-to-3x-faster\/\" rel=\"nofollow noopener\" target=\"_blank\">Multi-Token Prediction (MTP) drafters<\/a>, which take advantage of unused processing cycles to calculate possible future tokens. The result is greater speed and efficiency. Google has released optional MTP versions of the other Gemma 4 models, but this is the first one to have MTP out of the box.<\/p>\n<p>Gemma 4 12B is also more efficient thanks to a new approach to multimodality. The Gemma 4 family is natively multimodal, accepting text, audio, or images as inputs. Most gen AI models\u2014including the other Gemma 4 variants\u2014use dedicated encoders to process non-text inputs and pass that data to the LLM. This works well enough, but it increases latency and memory usage.<\/p>\n<p>With the new mid-weight model, Google has implemented a streamlined embedding module for vision, featuring single-matrix multiplication and positional embedding, which allows the data to pass to the LLM with proper spatial awareness. This eliminates the need for a bulky middleman encoder. For audio, there\u2019s no encoding at all. The developers worked out a method of projecting the raw audio signal into the same vectors used for text tokens.<\/p>\n<\/p>\n<p>If you want to check out the new Gemma 4 model, it\u2019s accessible without a download via tools like <a href=\"https:\/\/lmstudio.ai\/models\/gemma-4\" rel=\"nofollow noopener\" target=\"_blank\">LM Studio<\/a>, <a href=\"https:\/\/developers.google.com\/edge\/gallery\" rel=\"nofollow noopener\" target=\"_blank\">Google AI Edge Gallery<\/a>, and more. But the whole idea with Gemma 4 12B is that you can run it locally and on your own terms. If you\u2019ve got the RAM, the model weights are available for download immediately on <a href=\"https:\/\/huggingface.co\/collections\/google\/gemma-4\" rel=\"nofollow noopener\" target=\"_blank\">Kaggle<\/a> and <a href=\"https:\/\/huggingface.co\/collections\/google\/gemma-4\" rel=\"nofollow noopener\" target=\"_blank\">Hugging Face<\/a>. It\u2019s just shy of 18GB.<\/p>\n","protected":false},"excerpt":{"rendered":"Gemma 4 12B is almost as capable as the version with 26 billion parameters. Credit: Google Gemma 4&hellip;\n","protected":false},"author":2,"featured_media":654916,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[256,254,255,64,63,105],"class_list":["post-716603","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-au","tag-australia","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/716603","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/comments?post=716603"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/716603\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media\/654916"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media?parent=716603"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/categories?post=716603"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/tags?post=716603"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}