Gemini is Latin for “twins” and is the name of a constellation associated with Greek mythological twins Castor and Pollux, chosen partly because the Gemini zodiac sign represents a dual-natured personality. 1)
During early development, Gemini's placeholder name was “Titan,” after Saturn's largest moon, before the team settled on Gemini. 2)
Google merged its two separate AI research groups—Google Brain and DeepMind—in April 2023 to form Google DeepMind, and the combined team built Gemini together. 3)
Jeff Dean wrote a one-page memo arguing that it was “silly” for multiple teams to work separately toward the same goal, leading to the decision to combine people, ideas, and compute resources into one multimodal model. 4)
From the very beginning, Gemini was designed to understand audio, images, language, text, and other forms of information natively, rather than stitching together separate single-modality models. 5)
Gemini was lagging until its coding abilities were improved, which subsequently and unexpectedly improved the model's general reasoning ability across non-coding tasks as well. 6)
Gemini 1.0 was announced on December 6, 2023, as Google's first family of models designed to be natively multimodal, arriving in three sizes: Ultra, Pro, and Nano. 7)
On February 8, 2024, Google renamed its Bard chatbot to Gemini and launched Gemini Advanced, unifying the model family and the consumer assistant under a single name. 8)
Gemini 1.5, launched in February 2024, featured an experimental 1 million token context window, roughly equivalent to one hour of video, 11 hours of audio, or 700,000 words of text. 9)
Gemini 2.0, released in December 2024, was positioned around the “agentic era,” emphasizing the model's ability to autonomously perform multi-step tasks. 10)
Gemini 4 Argon's output token limit was expanded to an industry-leading 1 million tokens, up from the previous 64,000, enabling longer and more complex tasks in a single session. 11)
Gemini 4 Argon set a new state of the art on DeepSWE v1.1 with a score of 77.9%, surpassing Claude Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%. 12)
On CWE-bench v1, which evaluates a model's ability to remediate security vulnerabilities, Gemini 4 Argon tied for first place with a top score of 68%. 13)
On LVBench, which measures long video understanding, Gemini 4 Argon achieved state-of-the-art performance with a score of 91.7%. 14)
In one internal Google project, Gemini 4 Argon beat the published baseline for quantum algorithm optimization by 40% in a matter of minutes. 15)
A team of Argon agents analyzed fleet-wide profiling telemetry to autonomously identify and apply memory optimizations across Google's data centers, freeing up over 300 TiB of memory with an estimated 500 TiB to 1 PiB in total savings. 16)
Argon agents are working on migrating Google's C/C++ codebases to Rust, scaling from tens of thousands of lines in core libraries like re2 and libgav1 up to 800K+ lines for the Fuchsia Zircon kernel. 17)
For libgav1, Google's open-source video decoding software, Argon agents replaced 32K lines of SIMD code with safe Rust, producing a memory-safe decoder that runs 2.7x faster than the previous Rust port with identical video output. 18)
Gemini 4 Argon launched at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input token price. 19)
After the introductory period expires, Gemini 4 Argon's price will rise to $4 per million input tokens and $20 per million output tokens. 20)
Google cut the price of its top-tier AI Ultra subscription from $249.99 to $99.99 per month, its most aggressive subscription price move in the AI era. 21)
Gemini API access ranges from $0.10 per million input tokens for Flash-Lite to $2.00 per million for 3.1 Pro, with a free tier available in Google AI Studio for Flash models. 22)
Gemini's “thinking tokens”—the internal reasoning steps the model generates before responding—are billed at output token rates, meaning actual costs are higher than the base pricing page suggests. 23)
Gemini 3 Deep Think achieved 84.6% on ARC-AGI-2, a reasoning benchmark verified by the ARC Prize Foundation that tests a model's ability to adapt to entirely new tasks. 24)
Gemini 3 Deep Think set a new record by scoring 48.4% without tools on Humanity's Last Exam, a benchmark designed to test the absolute limits of frontier AI models. 25)
Gemini 3 Deep Think holds an Elo rating of 3455 on Codeforces, a competitive programming platform, and reached gold-medal-level performance on the 2025 International Math Olympiad. 26)
Beyond mathematics and coding, the updated Deep Think mode reached gold-medal-level results on the written sections of both the 2025 International Physics Olympiad and the Chemistry Olympiad. 27)
The updated Deep Think can analyze a hand-drawn design, model the complex geometry required to make it functional, and generate a 3D-printable file, opening doors for rapid prototyping. 28)
Gemini Omni, announced at Google I/O 2026, can combine images, audio, video, and text as input to generate high-quality videos grounded in Gemini's real-world knowledge. 29)
Gemini Omni allows users to edit videos through natural language conversation, with every instruction building on the last and characters and physics remaining consistent across edits. 30)
Omni has an improved intuitive understanding of forces like gravity, kinetic energy, and fluid dynamics, allowing users to create more realistic video scenes. 31)
Nano Banana 2.1, a member of the Gemini 3 series, is a natively multimodal reasoning model based on Gemini 3.6 Flash that can comprehend text and images and generate both image and text output. 32)
Nano Banana 2.1 has a token context window of up to 1 million tokens and generates images with a 4K token output and text with a 64K token output. 33)
Gemini's Personal Intelligence feature, launched as a beta in the U.S., securely connects information from apps like Gmail and Google Photos with a single tap to make Gemini uniquely helpful. 34)
Connecting apps through Personal Intelligence is off by default; users choose to turn it on, decide exactly which apps to connect, and can turn it off anytime. 35)
Gemini does not train directly on users' Gmail inboxes or Google Photos libraries; it trains on limited information like specific prompts and the model's responses to improve functionality. 36)
Gemini in Google Workspace can directly interact with Asana, Atlassian Rovo, HubSpot, Intuit Mailchimp, Intuit QuickBooks, Monday, and Salesforce through Model Context Protocol (MCP) integrations. 37)
Grounding with Google Search connects the Gemini model to real-time web content, allowing it to provide more accurate answers and cite verifiable sources beyond its knowledge cutoff. 38)
Gemini 4 Argon was first rolled out to trusted cyber defenders through Google's Fairwind Program, an initiative focused on cybersecurity defense, before broader release. 39)
Google announced in August 2026 that the Gemini app had more than one billion users per month, putting it in competitive position with OpenAI's ChatGPT. 40)
Gemini 3 was the first AI model to break 1500 Elo on LMArena, a crowdsourced benchmark for comparing language models. 41)
Gemini is integrated across Google's entire product ecosystem, powering Google Search, Workspace, Android, and Chrome by January 2026. 42)
In a 2024 controversy, Google's Gemini image generation tool produced historically inaccurate images, such as depicting non-White figures in Nazi-era contexts, due to a “diversity injection” mechanism that lacked contextual safeguards. 43)
In March 2026, the father of a Florida man sued Google in the first wrongful death case in the U.S. against the tech giant over alleged harms caused by Gemini, claiming the chatbot fueled a delusional spiral that led to his son's suicide. 44)
The lawsuit alleges Google made design choices that ensured Gemini would “never break character” so the firm could “maximise engagement through emotional dependency.” 45)
In November 2024, a Michigan graduate student received a threatening response from Gemini that said “You are not special, please die” while asking about challenges for older adults making income after retirement. 46)
Google launched Gemini Enterprise for Financial Services and Gemini Enterprise for Legal in August 2026, providing agentic capabilities tailored to financial work and extensive case-law research. 47)