Gemma

Gemma is Google's open-source model family, released in February 2024 alongside Gemini. While Gemini powers Google's consumer products, Gemma is designed for developers who need deployable, self-hostable models that run efficiently on limited hardware — laptops, phones, or small cloud instances.

**Model variants:** The Gemma family includes 2B and 7B parameter models (smaller versions optimized for speed) and the newer Gemma 2 series (2B, 9B, 27B) with significantly improved performance. CodeGemma is a specialized variant for programming tasks. PaliGemma handles vision-language tasks.

**Who it's for:** - Developers building AI features who want to avoid API costs at scale - Researchers needing a capable open model for experimentation - Companies with data privacy requirements that preclude sending data to external APIs - Edge computing applications where latency matters

**Running Gemma:** Models are available on Hugging Face, Google's Kaggle, and Vertex AI. Tools like Ollama, LM Studio, and llama.cpp support Gemma locally on consumer hardware. A 2B model runs comfortably on a modern laptop with 8GB RAM; the 7B variant needs 16GB.

**Performance:** Gemma 2 9B outperforms Llama 3 8B on many benchmarks and punches above its weight class. It's a capable model for summarization, question-answering, and light coding tasks — not frontier-tier, but competitive for its size.

**License:** Gemma uses the Gemma Terms of Use (not Apache 2.0). Commercial use is permitted but with restrictions around certain large-scale deployments.

**Alternatives:** Meta's Llama 3 is the most direct competitor with a more permissive license. Microsoft's Phi-3 family is similarly size-optimized. Mistral's models offer strong performance in the same tier.