Gemma

Available · Language · Milestone

Gemma is a family of lightweight open-weight language models from Google, released in February 2024 in 2B and 7B sizes, each pre-trained and instruction-tuned, and built from the research behind Gemini. The text-to-text, English-language models have an 8,192-token context window, and their terms of use allow responsible commercial use. [1] [2] [3] Primary source

Timeline of Gemma →

Claims and evidence

  • Released Primary source[1] [4]February 21, 2024
  • Status AvailableWeights still downloadable from Google's Hugging Face organisation (gated behind the Gemma terms) on 2026-10-01. Primary source[3] [2]
      • Input text Primary source[3]Inputs and outputs [2]Inputs and outputs
      • Output text Primary source[3]Inputs and outputs [2]Inputs and outputs
      • Feature Fine tuning availableGoogle provided toolchains and notebooks for supervised fine-tuning. Primary source[1]Optimized across frameworks, tools and hardware
      • Open weights Yes Primary source[1]Gemma open models [3] [2]
      • Context window 8192 tokens tokens Primary source[2]Context Length [5]section 2
      • Access Open-weights downloadWeights on Google's Hugging Face organisation (gated behind the Gemma terms); the announcement also names deployment on Vertex AI and Google Kubernetes Engine. Primary source[1]Gemma open models [3] [2]

      Lineage

      Predecessors

      No known predecessor.

      Successors

      Based on

      Not derived from another model.

      Variants and derived

      Siblings

      None recorded.

      All ancestors

      None.

      All descendants

      Variants

      Gemma 2B

      Same dates as Gemma.

      • Variant Inline variant in this record.Released as pre-trained and instruction-tuned (-it) checkpoints. Primary source[1]Gemma open models [3]

      Also known as: google/gemma-2b, google/gemma-2b-it

      Differs in:

      • Parameters: 2B Evidence not assessed [5]Table 2

      Gemma 7B

      Same dates as Gemma.

      • Variant Inline variant in this record.Released as pre-trained and instruction-tuned (-it) checkpoints. Primary source[1]Gemma open models [2]

      Also known as: google/gemma-7b, google/gemma-7b-it

      Differs in:

      • Parameters: 7B Evidence not assessed [5]Table 2

      Related AI Radar coverage

      AI Radar coverage starts in June 2026; no coverage linked yet.

      All model releases from Google on AI Radar →