Google has launched Gemma 4, a new family of open artificial intelligence models, expanding its open model range under an Apache 2.0 licence.<\/p>\n
The family comes in four sizes: Effective 2B, Effective 4B, 26B Mixture of Experts and 31B Dense. It is designed to run on a range of hardware, from mobile devices and laptops to developer workstations and accelerators.<\/p>\n
Since the first generation was introduced, Gemma has been downloaded more than 400 million times, with developers creating more than 100,000 variants, according to Google. The company positions the latest release as an advance in reasoning, coding, multimodal processing and support for longer context windows.<\/p>\n
The larger models can process up to 256K of context, while the smaller edge-focused models support 128K. All models can process images and video, while the E2B and E4B models also include native audio input for speech recognition and understanding.<\/p>\n
Model range<\/p>\n
According to Google, the 31B model is currently ranked third among open models on the Arena AI text leaderboard, while the 26B model ranks sixth. Both are aimed at researchers and developers seeking stronger reasoning performance on accessible hardware.<\/p>\n
Google says the unquantised bfloat16 versions of the 26B and 31B models can fit on a single 80GB Nvidia H100 GPU. Quantised versions can also run on consumer-grade GPUs for local uses such as coding assistants and automated workflows.<\/p>\n
The 26B Mixture of Experts model activates 3.8 billion of its total parameters during inference, a design intended to reduce latency. By contrast, the 31B Dense model targets users seeking stronger output quality and a base for fine-tuning.<\/p>\n
At the smaller end of the range, the E2B and E4B models are designed for phones, IoT devices and compact computing platforms. They were developed to limit memory use and battery drain while running fully offline with low latency on devices including smartphones, Raspberry Pi systems and Nvidia Jetson Orin Nano units.<\/p>\n
Open licence<\/p>\n
The Apache 2.0 licence is central to the release. It allows commercial use and modification with relatively few restrictions, which may make the models more attractive to developers and organisations that want to retain control over deployment and data handling.<\/p>\n
Google says the models are intended to give developers flexibility to deploy across on-premises and cloud environments. It also says they go through the same infrastructure security protocols as its proprietary systems.<\/p>\n
The launch reflects a broader competitive push in open-weight AI models, as companies try to balance performance with lower hardware requirements and easier local deployment. Interest in smaller models that can run on-device has grown as developers seek lower costs, reduced latency and greater control over privacy-sensitive applications.<\/p>\n
Developer focus<\/p>\n
Gemma 4 includes native support for function calling, structured JSON output and system instructions, features intended for software agents that interact with tools and application programming interfaces. Google also says the models were trained on more than 140 languages.<\/p>\n
Launch support covers a wide set of development frameworks and runtimes, including Hugging Face tools, vLLM, llama.cpp, MLX, Ollama and Nvidia software. Developers can also use the models in Android workflows and adapt them on platforms such as Colab, Vertex AI and local hardware.<\/p>\n
Google also highlighted previous work built on earlier Gemma models, including a Bulgarian language model called BgGPT created by INSAIT and work with Yale University on Cell2Sentence-Scale for cancer research. The examples are intended to show that open models can serve both local language applications and scientific use cases.<\/p>\n
In a crowded market, Google is trying to differentiate Gemma 4 by offering a range that spans mobile devices to high-end GPUs while keeping the models openly available. The emphasis on local deployment, multimodal input and permissive licensing may appeal to developers weighing open models against closed commercial systems.<\/p>\n
Google says Gemma 4 was built from the same research and technology base as Gemini 3.<\/p>\n","protected":false},"excerpt":{"rendered":"Google has launched Gemma 4, a new family of open artificial intelligence models, expanding its open model range…\n","protected":false},"author":2,"featured_media":367107,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[73693,8194,139124,157549,6160,159219,2800,53873,17617,367,11753,150825,19582,162232,111,139,69,192601,15748,620,71190,8695,145],"class_list":{"0":"post-367106","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-technology","8":"tag-ai-ethics-governance","9":"tag-ai-models","10":"tag-ai-security","11":"tag-application-programming-interface-api","12":"tag-artificial-intelligence-ai","13":"tag-developer-tools","14":"tag-edge-ai","15":"tag-energy-efficient","16":"tag-generative-ai-genai","17":"tag-google","18":"tag-hugging-face","19":"tag-internet-of-things-iot","20":"tag-large-language-models-llms","21":"tag-mobile-ai","22":"tag-new-zealand","23":"tag-newzealand","24":"tag-nz","25":"tag-ollama","26":"tag-open-source","27":"tag-openai","28":"tag-power-energy","29":"tag-raspberry-pi","30":"tag-technology"},"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/367106","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/comments?post=367106"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/367106\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media\/367107"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media?parent=367106"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/categories?post=367106"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/tags?post=367106"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}