Google DeepMind releases Gemma 4, its latest open-weights model
Google DeepMind released Gemma 4, a new family of open-weight models under Apache 2.0 license. Available now.
Google DeepMind released Gemma 4 today, a new family of open-weight models built on technology from Gemini 3. The weights are free to download, fine-tune, and deploy commercially under an Apache 2.0 license.
The family comes in four sizes, split across two different use cases.
The larger models (26B and 31B) are built for workstations and consumer GPUs. The 26B uses a Mixture of Experts architecture, meaning only about 4 billion parameters are active during inference, keeping memory requirements lower than a dense model of similar capability. The 31B is a full dense model aimed at tasks that need deeper reasoning.
The edge models (E2B and E4B) target mobile and IoT devices. Both support multimodal inputs — text, image, and audio — and are designed to run completely offline on hardware like phones, Raspberry Pi, and Jetson Nano.
- Context window: Up to 256K tokens, double what Gemma 3 supported
- Languages: 140+
- Key capabilities: Function calling, structured JSON output, agentic workflows, multimodal reasoning
Where to get it:
- Try it: Google AI Studio
- Download weights: Hugging Face, Ollama
The instruction-tuned variants of the 26B and 31B models are live now.
If you need on-demand GPUs for training, fine-tuning, inference, or running open-source models, give RunPod a try.
- Available hardware: H100, H200, A100, L40S, RTX 4090, RTX 5090, and 30+ more
- Cost: significantly cheaper than AWS or GCP, billed per second, no contracts
- Setup: spins up in under a minute, 30+ regions worldwide
Get the core business tech news delivered straight to your inbox. We track AI, automation, SaaS, and cybersecurity so you don't have to.
Just read what you want, and be done with it.
