Gemini

Google's entry into the public LLM race began under pressure from ChatGPT's November 2022 launch. In February 2023, CEO Sundar Pichai announced 'Bard' — an experimental conversational AI service powered by LaMDA (Language Model for Dialogue Applications). The announcement itself became a cautionary tale: a promotional demo video contained a factually incorrect answer about the James Webb Space Telescope, triggering a roughly $100 billion drop in Alphabet's market capitalization before the product had even launched publicly. Limited waitlist access opened on March 21, 2023 to US and UK users, and waitlist removal expanded to 180+ countries by May 2023. Early reviews were mixed: critics noted Bard was faster than ChatGPT but prone to bland, cautious, and sometimes inaccurate responses.

In April 2023, Google made a structural bet by merging Google Brain and DeepMind into a single entity — Google DeepMind — under CEO Demis Hassabis. The rationale was to combine DeepMind's fundamental research with Google Brain's product expertise to build more capable AI systems. The work that emerged from this consolidation was unveiled on December 6, 2023, when Pichai and Hassabis announced Gemini 1.0 at a virtual press conference. Unlike GPT-4, which was trained separately on text and vision, Gemini 1.0 was architected to be natively multimodal from the start — trained simultaneously on text, images, audio, video, and code. It shipped in three tiers: Gemini Ultra (complex tasks), Gemini Pro (general tasks, immediately integrated into Bard), and Gemini Nano (on-device use, launched on Pixel 8 Pro). Google co-founder Sergey Brin notably returned to the company to assist in developing the Gemini LLMs.

February 2024 marked the formal brand consolidation: Bard was renamed Gemini, the 'Duet AI' branding for Google Workspace was retired, and Gemini Advanced — powered by Ultra 1.0 — launched via the Google One AI Premium subscription at $19.99/month. That same month, Google also launched Gemini 1.5 Pro, introducing a breakthrough one-million-token context window enabled by a new mixture-of-experts architecture; at launch, the 1M-token window was a world record. Gemini 1.5 Flash debuted at Google I/O in May 2024, establishing the Flash tier as a speed-optimized, lower-cost alternative to Pro. However, February 2024 was also scarred by a high-profile controversy: users discovered that the Gemini image-generation feature was producing historically inaccurate depictions — showing people of color as Nazi soldiers and the Founding Fathers — while sometimes refusing prompts for white subjects. Google paused human image generation entirely, and Alphabet's market cap fell approximately $96.9 billion during the ensuing backlash.

Gemini 2.0 arrived in December 2024 with Gemini 2.0 Flash Experimental, introducing a Multimodal Live API for real-time audio and video, native image generation with SynthID watermarking, and expanded agentic capabilities including the experimental coding agent Jules for GitHub. By January 30, 2025, Gemini 2.0 Flash became the new default model; Gemini 2.0 Pro followed on February 5, 2025. March 2025 saw the release of Gemini 2.5 Pro, described by Google as its 'most intelligent model yet' — a reasoning model that pauses to think before answering, available through Google AI Studio. In June 2025, Google launched Gemini CLI, an open-source AI agent that brought Gemini's capabilities directly to the developer terminal with generous free usage limits. That same month, Google introduced a refreshed Gemini logo and brand identity.

The Gemini 3 generation launched in late 2025, with Gemini 3 Pro and its 'Deep Think' reasoning mode rolling out to Google AI Ultra subscribers, followed by Gemini 3 Flash on December 17, 2025 — which became the new default in the Gemini app. February 2026 brought Gemini 3.1 Pro, scoring 77.1% on ARC-AGI-2, more than double its predecessor, and characterized by Google as a major step in practical reasoning for science and engineering. On May 19, 2026, at Google I/O, Google launched Gemini 3.5 Flash — the first model in the new 3.5 family — positioning it as the current flagship. It outperforms Gemini 3.1 Pro on 11 of 15 published benchmarks, including Terminal-Bench 2.1 (76.2%), GDPval-AA (1,656 Elo), and MCP Atlas (83.6%), runs at 280+ output tokens per second, and is priced at $1.50/$9 per million input/output tokens. Also announced at I/O 2026 were Gemini Omni (any-input-to-video model), Gemini Spark (persistent personal agent), Antigravity 2.0 (agentic dev platform), and Managed Agents in the Gemini API.