Google DeepMind has officially launched Gemma 4, a suite of open-weight AI models, following a viral leak on the LMSYS Chatbot Arena. The models, ranging from 2B to 27B parameters, prioritize speed and local deployment. Developers discovered the identity after the model, codenamed "significant-otter," confirmed its own origin during testing.
Google DeepMind released Gemma 4 today as a series of open-weight models designed for high-speed performance on consumer hardware. This official launch follows a week of intense speculation triggered by a mysterious "significant-otter" model appearing on the LMSYS Chatbot Arena.
Visitors to the LMSYS Chatbot Arena noticed the unlisted model handling complex benchmarks with unusual efficiency and stable output. When asked about its identity, the AI directly identified itself as Gemma 4, a large language model developed by Google DeepMind.
Google DeepMind has optimized these new versions to run on consumer hardware like laptops and mobile devices without requiring expensive cloud APIs. The lineup includes 2B, 9B, and 27B parameter variants that support over 140 languages and native multimodal processing for interactive assistants.
Developers building interactive assistants can now deploy these models locally to ensure data privacy and eliminate the recurring costs of external server calls. This level of accessibility positions Gemma 4 as a practical tool for real-time applications where a slow answer is as good as no answer.
Primary Evidence: https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/
Discussion