Open-model testing made easier with Gemma 4
Microsoft announced that Gemma 4 is now available in Microsoft Foundry, signalling easier experimentation with open models inside managed platforms. The announcement frames managed access to an open model as part of a broader push to make model evaluation routine for product teams. (techcommunity.microsoft.com)
A model is the engine behind a chatbot, and Microsoft said on April 14 that Google DeepMind’s Gemma 4 can now be tested and deployed inside Microsoft Foundry. (techcommunity.microsoft.com) Gemma 4 is an open-weights model family, which means developers can inspect and run the released model weights on their own hardware instead of using only a closed application programming interface. Google said the family includes Effective 2B, Effective 4B, 26B Mixture of Experts, and 31B Dense variants. (ai.google.dev, blog.google) Google said Gemma 4 handles text and image input, supports more than 140 languages, and offers context windows from 128,000 to 256,000 tokens, which is the amount of text a model can keep in working memory at once. Microsoft said customers can access the models in Foundry through the Hugging Face collection. (ai.google.dev, ai.google.dev, techcommunity.microsoft.com) Microsoft Foundry is the company’s managed catalog for finding, evaluating, and deploying artificial intelligence models inside Azure. Microsoft’s documentation says the catalog spans models from Azure OpenAI, Mistral, Meta, Cohere, NVIDIA, Hugging Face, and Microsoft-trained systems. (learn.microsoft.com, learn.microsoft.com) That matters for product teams because model testing usually breaks across different tools, contracts, and safety settings when one team compares closed models with open ones. Microsoft said Foundry gives Azure customers a single control plane, and in this announcement it said developers can use the same policies they already use for other workloads. (techcommunity.microsoft.com, learn.microsoft.com) Microsoft used the launch to argue that open-model access belongs inside the same enterprise workflow as proprietary models. The company said Foundry is the only hyperscaler platform offering OpenAI, Anthropic, Gemma, and more than 11,000 models under one control plane. (techcommunity.microsoft.com) Google has been pushing Gemma as the open side of a two-track strategy alongside its Gemini line of proprietary models. In its April 2026 launch post, Google said Gemma 4 was built for advanced reasoning and agentic workflows, with smaller versions aimed at mobile, browser, and edge use. (blog.google, ai.google.dev) The setup still leaves customers responsible for how they use the model. Microsoft’s Foundry documentation says customers must review provider model cards, choose models that fit their use case, and add safeguards such as Azure AI Content Safety to meet Microsoft’s product terms and code of conduct. (learn.microsoft.com) The immediate change is simple: a Google open model that launched in early April is now available inside Microsoft’s managed platform less than two weeks later. That gives Azure customers one more way to compare open and closed systems without leaving the same testing and deployment stack. (blog.google, techcommunity.microsoft.com)
