Gemma 2 27B IT

Gemma 2 27B IT is Google DeepMind's largest Gemma 2 model, a 27-billion-parameter instruction-tuned transformer released July 2024 and distilled from Gemini Ultra training data. It posts competitive MMLU and MT-Bench scores for its parameter class and handles instruction-following reliably in single-turn and short-context settings. The 8K context ceiling is the hard constraint: anything requiring long document windows, multi-turn memory, or RAG over large corpora should go to Llama 3.1 70B or Qwen 2.5 72B instead. The Gemma license is permissive for commercial use but is not OSI-approved, so check your legal team's open-source policy before deploying. Expect this tier to lose ground to longer-context successors as the Gemma line evolves through 2026.