Gemini

Google's generative-AI product effort began as a direct, defensive response to OpenAI's ChatGPT, which launched in November 2022 and immediately went viral. Internally, Google management declared a 'code red'; co-founders Larry Page and Sergey Brin, long retired from day-to-day roles, attended emergency meetings to assess the competitive threat, and Brin reportedly accessed the codebase for the first time in years. On February 6, 2023, CEO Sundar Pichai announced Bard — a chatbot powered by Google's LaMDA language model — to a group of trusted testers. The launch was immediately marred when a promotional demo clip showed Bard answering a question about the James Webb Space Telescope incorrectly; Alphabet's market capitalization fell roughly $100 billion overnight. Bard opened to the general public in March 2023, receiving mixed reviews for speed and creativity alongside concerns about factual accuracy.

On December 6, 2023, Pichai and Google DeepMind CEO Demis Hassabis jointly announced Gemini 1.0 — the company's first model explicitly designed as a natively multimodal architecture trained on text, code, images, audio, and video simultaneously. The 1.0 family shipped in three tiers: Gemini Ultra (for highly complex tasks), Gemini Pro (general-purpose), and Gemini Nano (on-device). Gemini Ultra was the first language model to outperform human experts on the 57-subject Massive Multitask Language Understanding benchmark, scoring 90 percent. Gemini Pro was integrated into Bard immediately, while Ultra was held for safety testing. In February 2024, Google completed the product rebrand: on February 8, Bard and Duet AI were unified under the Gemini brand, a Gemini mobile app launched on Android, and 'Gemini Advanced with Ultra 1.0' was released through a paid 'Google One AI Premium' subscription. That same month, Gemini 1.5 introduced a new Mixture-of-Experts architecture and a one-million-token context window — then a world record — alongside the launch of Gemma, Google's open-weight model line.

February 2024 was also the month Gemini's image generation feature triggered one of the most visible AI controversies of the era. Shortly after the feature launched, users discovered the model generating historically inaccurate images of people — including racially diverse Nazi-era German soldiers, Black female popes, and multiethnic U.S. Founding Fathers. Google paused people-image generation entirely on February 22, 2024; CEO Sundar Pichai called the outputs 'completely unacceptable' in an internal memo. Engineers had applied diversity-correction tuning that failed to distinguish prompts where historical specificity made such adjustments inappropriate. SVP Prabhakar Raghavan later published a blog post acknowledging the model had 'overcompensated.' The feature was eventually restored with revised guardrails, but the incident became a defining case study in AI alignment failure and damaged user trust at a critical moment in the competitive AI race.

The 2025 product cycle marked Google's deliberate pivot toward reasoning and agentic AI. Gemini 2.0 Flash became the default model on January 30, 2025, introducing native multimodal outputs and a Multimodal Live API for real-time audio and video interaction. On March 25, 2025, Gemini 2.5 Pro Experimental launched as Google's first explicitly designated 'thinking model,' capable of chain-of-thought reasoning before responding; at debut it reached the top of LMArena and led benchmarks including GPQA Diamond (84%) and Humanity's Last Exam (18.8%). Gemini 2.5 Pro and Flash reached general availability on June 17, 2025. In August 2025, an image-generation model codenamed 'Nano Banana' (officially Gemini 2.5 Flash Image) launched and became a viral internet sensation for its photorealistic 3D-figurine outputs. On November 18, 2025, Google unveiled Gemini 3 Pro as a 'reasoning-first' flagship with a Deep Think mode — achieving 91.9 percent on GPQA Diamond and 41 percent on Humanity's Last Exam without tools — alongside the new Google Antigravity agentic development platform. Gemini 3 Flash followed in December 2025 as the new default across the Gemini app and AI Mode in Google Search.

In early 2026 the pace intensified. Apple announced in January 2026 plans to integrate Gemini into an upcoming version of Siri. Gemini 3.1 Pro released February 19, 2026, posting 94.3 percent on GPQA Diamond and 80.6 percent on SWE-Bench Verified, described as a 2× reasoning leap over its predecessors. At Google I/O on May 19, 2026, Google unveiled the Gemini 3.5 family, launching with Gemini 3.5 Flash — the first Flash-tier model to outperform the previous-generation Pro on coding and agentic benchmarks — running roughly 4× faster than comparable frontier models at $1.50 per million input tokens. I/O 2026 also introduced Gemini Omni (a video-generation world model), Gemini Spark (a persistent personal AI agent for consumers, built on 3.5 Flash), Managed Agents in the Gemini API, and Antigravity 2.0. Gemini 3.5 Pro was announced for release in June 2026. By November 2025, Google had reported that the Gemini app surpassed 650 million monthly active users and AI Overviews in Search reached 2 billion monthly users.