AI Radar

Gemma 3n

Available · Language, Multimodal

Gemma 3n is a family of open-weight models from Google for on-device use, previewed on 20 May 2025 and fully released on 26 June 2025 in E2B and E4B sizes. Google presents it as a mobile-first architecture developed with Qualcomm, MediaTek and Samsung System LSI; it accepts text, image, audio and video input and returns text. [1] [2] [3]Inputs and outputs Primary source

Timeline of Gemma 3n →

Claims and evidence

  • Announced Primary source[1] [4]May 20, 2025
  • Released Full release with open weights; an early preview was available from 2025-05-20. Primary source[2] [5]June 26, 2025
  • Preview in Google AI Studio, the Gemini API and Google AI Edge Primary source[1]Get Started: Preview Gemma 3n Today [4]May 20, 2025
  • Status AvailableWeights remain downloadable from Google's Hugging Face organisation (checked 2026-10-01). Primary source[6] [3]Resources and Technical Documentation
    • Change · Architecture New mobile-first architecture with MatFormer nesting, Per-Layer Embeddings and KV cache sharing.Compared with Gemma 3 Primary source[2]What's new in Gemma 3n?
    • Change · Modality Adds audio input, for speech recognition and speech translation, next to text, image and video.Compared with Gemma 3 Primary source[2]
    • Change · Efficiency Google reports faster responses on mobile and a smaller memory footprint than Gemma 3 4B.Compared with Gemma 3 Primary source[1]
    • Input text, image, audio, video Primary source[3]Description; Inputs and outputs [2]What's new in Gemma 3n?
    • Output text Primary source[3]Inputs and outputs
    • Feature Multilingual140 languages for text and multimodal understanding of 35 languages. Primary source[2]What's new in Gemma 3n?
    • Open weights YesPre-trained and instruction-tuned weights under the Gemma Terms of Use; access on Hugging Face requires accepting the licence. Primary source[6] [3]Description
    • Context window 32K tokens Primary source[3]Inputs and outputs
    • Access Open-weights download, On device, APIDesigned to run on phones, tablets and laptops; gemma-3n-e4b-it is also offered in Google AI Studio and the Gemini API. Primary source[2] [1]Get Started: Preview Gemma 3n Today [4]May 20, 2025

    Lineage

    Predecessors

    No known predecessor.

    Successors

    No known successor.

    Based on

    Not derived from another model.

    Variants and derived

    None recorded.

    Siblings

    None recorded.

    All ancestors

    None.

    All descendants

    None.

    Variants

    Gemma 3n E2B

    Same dates as Gemma 3n.

    • Variant Inline variant in this record. Primary source[2]What's new in Gemma 3n? [5]June 26, 2025

    Also known as: gemma-3n-E2B-it

    Differs in:

    • Parameters: 5B Evidence not assessed [2]What's new in Gemma 3n?

    Gemma 3n E4B

    Same dates as Gemma 3n.

    • Variant Inline variant in this record. Primary source[2]What's new in Gemma 3n? [6]

    Also known as: gemma-3n-e4b-it

    Differs in:

    • Parameters: 8B Evidence not assessed [2]What's new in Gemma 3n?

    Related AI Radar coverage

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

    All model releases from Google on AI Radar →