Welcome to
the Gemini era

  • The Gemini era
  • Capabilities
  • Hands-on
  • Safety
  • Bard
  • Build with Gemini
  • The Gemini ecosystem represents Google's most capable AI.

    Gemini is built from the ground up for multimodality — reasoning seamlessly across text, images, video, audio, and code.

    The Gemini era

    Gemini represents a significant leap forward in how AI can
    help improve our daily lives.

    Meet the first version of Gemini
    our most capable AI model.

    89.8%

    Gemini Ultra
    90.0%
    CoT@32*

    Human expert (MMLU)

    86.4%

    5-shot* (reported) Previous SOTA (GPT-4)

    *Note that evaluations of previous SOTA models
    use different prompting techniques.

    Gemini is the first model to
    outperform human experts on
    MMLU (Massive Multitask
    Language Understanding), one of
    the most popular methods to test
    the knowledge and problem
    solving abilities of AI models.

    TEXT

    Capability Benchmark
    Higher is better
    Description Gemini Ultra GPT-4
    API numbers calculated where reported numbers were missing
    General MMLU Representation of questions in 57 subjects (incl.
    STEM, humanities, and others)
    90.0%
    CoT@32*
    86.4%
    5-shot*(reported)
    Reasoning Big-Bench Hard

    DROP

    HellaSwag
    Diverse set of challenging tasks requiring multi-step
    reasoning


    Reading comprehension (F1 Score)

    Commonsense reasoning for everyday tasks
    83.6%
    3-shot

    82.4
    Variable shots

    87.8%
    10-shot*
    83.1%
    3-shot(API)

    80.9
    3-shot(reported)

    95.3%
    10-shot(reported)
    Math GSM8K

    MATH
    Basic arithmetic manipulations (incl. Grade School
    math problems)


    Challenging math problems (incl. algebra, geometry,
    pre-calculus, and others)
    94.4%
    MAJ1@32

    53.2%
    4-shot(API)
    92.0%
    5-shot CoT(reported)

    52.9%
    4-shot(API)
    Code HumanEval

    Natural2Code
    Python code generation

    Python code generation. New held out dataset
    HumanEval-like, not leaked on the web
    74.4%
    0-shot(IT)*

    74.9%
    0-shot
    67.0%
    0-shot*(reported)

    73.9%
    0-shot(API)

    *See the technical report for details on performance with other methodologies

    Gemini surpasses SOTA performance on
    all multimodal tasks.

    MULTIMODAL

    Capability Benchmark
    Higher is better
    Description Gemini Ultra GPT-4
    API numbers calculated where reported numbers were missing
    Image
    MMLU

    MMLU

    MMLU

    MMLU
    Representation of questions in 57 subjects (incl.
    STEM, humanities, and others)
    90.0%
    CoT@32*
    86.4%
    5-shot*(reported)
    Audio GSM8K

    MATH
    Basic arithmetic manipulations (incl. Grade School
    math problems)


    Challenging math problems (incl. algebra, geometry,
    pre-calculus, and others)
    94.4%
    MAJ1@32

    53.2%
    4-shot(API)
    92.0%
    5-shot CoT(reported)

    52.9%
    4-shot(API)
    Video HumanEval

    Natural2Code
    Python code generation

    Python code generation. New held out dataset
    HumanEval-like, not leaked on the web
    74.4%
    0-shot(IT)*

    74.9%
    0-shot
    67.0%
    0-shot*(reported)

    73.9%
    0-shot(API)

    *See the technical report for details on performance with other methodologies

    Gemini comes in three sizes

    Ultra

    Pro

    Nano

    Our most capable and
    largest model for highly-
    complex tasks.

    Our best model for scaling
    across a wide range of
    tasks.

    Our most efficient model
    for on-device tasks.

    Anything to anything

    Gemini is natively multimodal, which gives you the
    potential to transform any type of input into any
    type of output.

    Hands-on with our Gemini model

    Watch highlights from testing of our Gemini
    model's multimodal reasoning capabilities. Curious
    to learn more? Explore our prompting techniques here.

    I.

    Multimodal Dialogue

    II.

    Multilinguality

    III.

    Game Creation

    IV.

    Visual Puzzles

    V.

    Making Connections

    The potential of Gemini

    Learn about what our Gemini models can do from
    some of the people who built it.

    RÉMI LEBLOND AND GABRIELA SURITA

    Excelling at competitive programming

    TAYLOR APPLEBAUM AND SEBASTIAN NOWOZIN

    Unlocking insights in scientific literature

    ADRIÀ RECASENS

    Processing and understanding raw audio signal end-to-end

    SAM CHEUNG

    Explaining reasoning in math and physics

    PALASH NANDY

    Reasoning about user intent to generate bespoke experiences

    Building and deploying
    Gemini responsibly

    We've built our Gemini models responsibly from the start,
    incorporating safeguards and working together with
    partners to make it safer and more inclusive.

    Try Gemini Advanced with
    our most capable AI model

    With Ultra 1.0, Gemini Advanced is far more capable at coding, reasoning, and creative collaboration.


    Learn More

    Build with Gemini

    Integrate Gemini models into your applications with Google AI Studio
    and Google Cloud Vertex AI.

    See more what #GeminiAI
    can do