Google DeepMind has rolled out Gemma 4, a fresh lineup of open AI models built for demanding reasoning tasks and self-directed AI workflows. The launch marks another step forward in Google’s bid to grow its footprint in the open-model space, putting more powerful tools in the hands of developers under the permissive Apache 2.0 open-source licence.
The Gemma platform has gained a lot of popularity since it was first launched. Developers have downloaded the original Gemma models more than 400 million times and have built over one lakh distinct variations from them, Google said.
Google Gemma 4: Features and Capabilities
Gemma 4 ships in four configurations: Effective 2B, Effective 4B, a 26B Mixture of Experts model, and a 31B Dense model. The range is built to span the full hardware spectrum — from a smartphone or a laptop graphics card all the way up to enterprise-grade AI servers — giving developers flexibility depending on what they are building and where it needs to run.
Reasoning is one of the headline improvements in Gemma 4. The models handle multi-step problems with greater accuracy than earlier versions and show stronger results on mathematics benchmarks and tasks that require precise instruction-following.
For developers building AI agents, Gemma 4 brings native support for function calling, structured JSON output, and system-level instructions. This makes it considerably more straightforward to wire up an AI agent that talks to external tools, third-party APIs, and other services. The models are also capable of producing clean, reliable code without needing an internet connection.
Context window capacity has also grown. Smaller edge-focused models support up to 128,000 tokens, while the larger models stretch to 256,000 tokens. Gemma 4 also handles more than 140 languages, broadening its usefulness for developers building products for international audiences.
Where to Access Gemma 4:
The 31B Dense and 26B MoE models are live on Google AI Studio. Developers looking to work with the lighter E4B and E2B variants can find them through the Google AI Edge Gallery.
Google launches AI Pro subscription plan:
Google is making a more aggressive move into the paid AI space with the launch of its AI Pro subscription plan. The offering takes aim at users who lean on AI tools as part of their daily routine — professionals, students, and content creators among them. Rather than scattering features across separate products, Google has bundled its advanced AI capabilities into a single plan, with the goal of making routine tasks quicker and more straightforward across its wider product ecosystem.