Gemini 2.0 Flash
Google DeepMind's fast multimodal workhorse model for agentic apps
Gemini 2.0 Flash is a multimodal language model from Google DeepMind, first shown in December 2024 as the opening member of the Gemini 2.0 line. It handles text, images, video and audio, and can also produce images and speech, with built-in tool calling aimed at assistant-style agents.
- First model in the Gemini 2.0 family, shown in December 2024
- Google DeepMind says it beats 1.5 Pro on key benchmarks at twice the speed
- Outputs include generated images and steerable multilingual speech
- Calls Google Search, code execution and third-party functions natively
- Trained and served wholly on Google TPUs, including sixth-generation Trillium
What it is
Gemini 2.0 Flash is a general-purpose multimodal model built by Google DeepMind. It arrived in December 2024 as an experimental release and the first entry in the Gemini 2.0 family. The company positions it as its fast, everyday workhorse: quick to respond, but stronger than earlier Flash versions. Beyond reading text, pictures, video and sound, it can write back with generated images and with spoken audio in several languages. It can also trigger tools such as web search and code execution, which is what makes it useful as the engine behind assistant software that acts on a user's behalf.
How it came about
Google DeepMind announced the model in December 2024, with CEO Demis Hassabis and CTO Koray Kavukcuoglu writing on behalf of the Gemini team. It follows Gemini 1.0, launched the previous December, and the 1.5 series. The company says it had been circulating early experimental builds to developers for about a month before the public announcement. A chat-tuned build went live for Gemini users worldwide on desktop and mobile web that same day, with the mobile app to follow. General availability and additional model sizes were promised for January 2025.
Inside the model
The company describes 2.0 Flash as a faster successor to 1.5 Flash, its most popular developer model, keeping similar response times while raising capability. Inputs cover text, images, video and audio; outputs add natively generated pictures mixed with text and steerable multilingual text-to-speech. Tool calling is built in, covering Google Search, code execution and user-defined third-party functions. Training and inference ran entirely on Google's own accelerator chips, including the sixth-generation TPU called Trillium. The company has not published parameter counts, training data volumes or layer details for this model.
What it can do
Google DeepMind claims 2.0 Flash beats the larger 1.5 Pro on key benchmarks while running at twice the speed, though it does not give the scores. It cites multimodal reasoning, long-context understanding, instruction following, planning and compositional function calling among its strengths. A Multimodal Live API was released alongside it for real-time audio and video-streaming input with several tools used together. The company frames these traits as the basis for agentic prototypes such as Project Astra, Project Mariner and the coding agent Jules.
How to use it
Developers reached the experimental model through Google AI Studio and Vertex AI. Multimodal input with text output was open to all developers, while speech synthesis and native image generation were limited to early-access partners. Gemini users could pick a chat-optimised build from the model menu on desktop and mobile web, with the mobile app to come later. General availability was scheduled for January 2025 alongside further model sizes.
Safety and testing
Google states that safety and security are central priorities for the Gemini 2.0 programme and that it aims to put models in users' hands quickly but responsibly. Feedback from trusted testers of Project Astra, running on Android phones, is described as informing work on the safety and ethics of a universal assistant. The company gives no benchmark safety results for 2.0 Flash itself.
Reception and debate
The company reports strong developer feedback during the month of pre-release testing but publishes no independent assessments. It presents the model as the foundation for a new phase of assistant products, including AI Overviews in Search, which it says reach one billion people. No third-party reviews, disputes or restrictions are described.
Why it matters
Google DeepMind positions 2.0 Flash as the base for its agentic direction, powering prototypes that see, plan and act: Project Astra for a universal assistant, Project Mariner for browser control, and Jules for software work. The company says Gemini models already run across seven of its products used by two billion people, and that 2.0 capabilities would reach more of them early the following year. Trillium TPUs, used for the whole training and inference pipeline, became generally available to cloud customers at the same time.
Questions people ask about Gemini 2.0 Flash
What is Gemini 2.0 Flash?
A multimodal language model from Google DeepMind, released experimentally in December 2024 as the first member of the Gemini 2.0 family. It is designed as a quick, general-purpose model that can also generate images and speech and call external tools.
How is it different from Gemini 1.5 Flash?
Google DeepMind says it keeps the fast response times of 1.5 Flash while raising capability, and claims it surpasses 1.5 Pro on key benchmarks at twice the speed. It also adds native image and audio output plus built-in tool calling, which the earlier Flash line lacked.
How can I use it?
Developers accessed the experimental build through Google AI Studio and Vertex AI, with text output from multimodal input open to all and speech and image generation reserved for early-access partners. Gemini users could select a chat-tuned version on desktop and mobile web. General availability was set for January 2025.
How large is it?
Google DeepMind has not disclosed a parameter count for Gemini 2.0 Flash. It describes the model only in terms of speed and capability relative to 1.5 Flash and 1.5 Pro.
What hardware trains and runs it?
The company says every part of training and inference for Gemini 2.0 used its own tensor processing units, including the sixth-generation chip named Trillium, which became generally available to cloud customers at the time of the announcement.
Good reading and sources
There is no English Wikipedia article on Gemini 2.0 Flash yet, so the facts on this page come from the sources listed above: blog.google/technology/google-deepmind/google-gemini-ai-update-decembe. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.