https://deepmind.google/technologies/gemini/ Jump to Content DeepMind Search [ ] Search Close DeepMind * About + Overview + Responsibility & Safety + Education + Careers + Press * Technologies * Impact * Discover + Overview + Blog + The Podcast + Visualising AI Search [ ] Search Close Welcome to the Gemini era * * The Gemini era * Capabilities * Hands-on * Safety * Bard * Build with Gemini * 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. Read the blog post Meet the first version of Gemini -- our most capable AI model. Gemini Ultra 90.0% 90.0% CoT@32* 89.8% Human expert (MMLU) 86.4% 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 GPT-4API numbers Benchmark calculated TEXT Capability Higher is Description Gemini Ultra where better reported numbers were missing MMLURepresentation Representation of questions in 57 of questions in General subjects (incl. 57 subjects 90%CoT@32* 86.4%5-shot* STEM, humanities, (incl. STEM, (reported) and others) humanities, and others) Big-Bench Hard Diverse set of Diverse set of challenging 83.1%3-shot Reasoning challenging tasks tasks requiring 83.6%3-shot (API) requiring multi-step multi-step reasoning reasoning DROPReading Reading 82.4Variable 80.93-shot comprehension (F1 comprehension shots (reported) Score) (F1 Score) HellaSwagCommonsense Commonsense 95.3%10-shot* reasoning for reasoning for 87.8%10-shot* (reported) everyday tasks everyday tasks GSM8KBasic Basic arithmetic arithmetic Math manipulations (incl. manipulations 94.4%maj1@32 92%5-shot CoT Grade School math (incl. Grade (reported) problems) School math problems) MATHChallenging math Challenging problems (incl. math problems algebra, geometry, (incl. algebra, 53.2%4-shot 52.9%4-shot pre-calculus, and geometry, (API) others) pre-calculus, and others) Code HumanEvalPython code Python code 74.4%0-shot 67%0-shot* generation generation (IT)* (reported) Python code Natural2CodePython generation. New code generation. New held out 73.9%0-shot held out dataset dataset 74.9%0-shot (API) HumanEval-like, not HumanEval-like, leaked on the web not leaked on the web *See the technical report for details on performance with other methodologies Show all (8) Gemini surpasses SOTA performance on all multimodal tasks. MULTIMODAL GPT-4VPrevious Description SOTA model MULTIMODAL Capability Benchmark Higher is better Gemini listed when unless otherwise capability is noted not supported in GPT-4V MMMUMulti-discipline Multi-discipline 59.4%0-shot 56.8%0-shot Image college-level college-level pass@1 pass@1 reasoning problems reasoning Gemini Ultra GPT-4V problems (pixel only*) VQAv2Natural image Natural image 77.8%0-shot 77.2%0-shot understanding understanding Gemini Ultra GPT-4V (pixel only*) TextVQAOCR on natural OCR on natural 82.3%0-shot 78%0-shot images images Gemini Ultra GPT-4V (pixel only*) DocVQADocument Document 90.9%0-shot 88.4%0-shot understanding understanding Gemini Ultra GPT-4V (pixel only*) (pixel only) Infographic VQA Infographic 80.3%0-shot 75.1%0-shot Infographic understanding Gemini Ultra GPT-4V understanding (pixel only*) (pixel only) MathVistaMathematical Mathematical 53%0-shot 49.9%0-shot reasoning in visual reasoning in Gemini Ultra GPT-4V contexts visual contexts (pixel only*) VATEXEnglish video English video 62.74-shot 564-shot Video captioning captioning Gemini Ultra DeepMind (CIDEr) (CIDEr) Flamingo Perception Test MCQA Video question 54.7%0-shot 46.3%0-shot Video question answering Gemini Ultra SeViLA answering CoVoST 2 (21 Automatic speech Audio languages)Automatic translation 40.1Gemini Pro 29.1Whisper speech translation (BLEU score) v2 (BLEU score) FLEURS (62 languages) Automatic speech Automatic speech recognition recognition (based on word 7.6%Gemini Pro 17.6%Whisper (based on word error error rate, v3 rate, lower is lower is better) better) *Gemini image benchmarks are pixel only--no assistance from OCR systems Show all (10) Read the technical report Gemini comes in three sizes Ultra Our most capable and largest model for highly-complex tasks. Pro Our best model for scaling across a wide range of tasks. Nano 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. Following content is a visual/ descriptive representation of the functionality of Gemini: Gemini can generate code based on different inputs you give it. Geminicangeneratecodebasedondifferentinputsyougiveit. Could Gemini help make a demo based on this video? Gemini I see a murmuration of starlings, so I coded a flocking simulation. class Boid { constructor(x, y) { this.pos = new p5.Vector(x, y); this.vel = p5.Vector.random2D(); this.vel.setMag(random(2, 4)); this.acc = new p5.Vector(); this.maxForce = 0.2; this.maxSpeed = 4; } } Gemini can generate text and images, combined. Geminicangeneratetextandimages,combined. Could Gemini show me ideas for what to make? Gemini Pink and blue mouse octopus knit how about an octopus with blue and pink tentacles? Pink and blue mouse knit how about an octopus with blue and pink tentacles? Gemini can reason visually across languages. Geminicanreasonvisuallyacrosslanguages. User has uploaded a voice prompt asking 'Explain how I can play this', and uploaded a sheet of music. Could Gemini explain what this means? Gemini I see the time signature is 6/8. This means there are 6 eighth notes in each measure. The dynamic marking is piano, which means to play softly. Andante grazioso means to play at a graceful walking pace. Hands-on with Gemini Watch highlights from our testing of Gemini's multimodal reasoning capabilities. Curious to learn more? Explore our prompting techniques here. * Multimodal Dialogue * Multilinguality * Game Creation * Visual Puzzles * Making Connections * Image & Text Generation * Logic & Spatial Reasoning * Translating Visuals * Cultural Understanding The potential of Gemini Learn about what Gemini can do from some of the people who built it. Read the blog post Image: two people standing at a table about to say something. Taylor Applebaum and Sebastian Nowozin Unlocking insights in scientific literature Image: two people standing at a table with a computer in front of a curtain. Remi Leblond and Gabriela Surita Excelling at competitive programming Image: a person with glasses standing in a room, smiling and ready to speak. Adria Recasens Processing and understanding raw audio signal end-to-end Image: a person with glasses sitting in front of a computer speaking. Sam Cheung Explaining reasoning in math and physics Image: a person with a beard standing in front of an open comptuer, similing and ready to speak. Palash Nandy Reasoning about user intent to generate bespoke experiences Building and deploying Gemini responsibly We've built Gemini responsibly from the start, incorporating safeguards and working together with partners to make it safer and more inclusive. Bringing Gemini Pro to Bard Experience Gemini Pro in Bard and unlock new ways to create, plan, brainstorm, and more. bard.google.com Line background Build with Gemini Integrate Gemini models into your applications with Google AI Studio and Google Cloud Vertex AI. Available December 13th. Coming soon See more of what #GeminiAI can do * Can Gemini turn images into code? * Can Gemini find similarities between two images? * Can Gemini understand unusual emojis? * Can Gemini understand outfits? * Can Gemini guess the movie? * Can Gemini make sense of my environment? Load More Footer links Follow us * * * * * Email (e.g., name@example.com) [ ] Please enter a valid email (e.g., "name@example.com") Thank you for signing up Something went wrong, please try again. I accept Google's Terms and Conditions and acknowledge that my information will be used in accordance with Google's Privacy Policy. Sign up * About Google * Google products * Privacy * Terms * Cookies management controls