Gemma 2 2B IT, Gemma 2 9B IT, and Gemma 2 27B IT are Google DeepMind's full Gemma 2 lineup, all released July 2024 and all carrying the same 8K context ceiling. All three are distilled from Gemini Ultra training data and share the Gemma license, which is permissive for commercial use but not OSI-approved — worth checking with legal before production deployment if your organization requires OSI-compliant licenses. The 8K context ceiling is the structural constraint that defines the entire family. For any workload requiring document-length inputs, multi-turn memory, or RAG over large corpora, all three models require chunking or an alternative. Teams building long-context applications should instead evaluate Llama 3.1 70B, Qwen 2.5 72B, or Phi-3 Medium before committing to Gemma 2. The 2B model targets edge and on-device inference. On classification and named-entity extraction it performs comparably to Llama 3.2 3B and IBM Granite 2B, at minimal compute cost. The economics typically favor self-hosting on cheap GPU or on-device deployment over hosted API usage at this scale. The 9B is the most practically deployed tier. It benchmarks closely with Llama 3.1 8B and Mistral 7B v0.3 on standard evals, and Groq hosts it with some of the lowest latency numbers available at sub-10B scale. Pricing across providers is competitive for high-throughput classification and short-form generation workloads. The 27B is the quality ceiling of the family, distilled from Gemini Ultra and posting competitive MT-Bench and MMLU scores for its parameter class. The 8K context remains the hard ceiling, but for single-turn instruction-following within that window it delivers reliable results. Pick the 2B for edge deployment or on-device inference. Pick the 9B for high-throughput hosted workloads that fit within 8K context. Pick the 27B when quality within short-context tasks is the priority.