Gemini

Gemini's lineage begins with Bard, Google's conversational assistant that opened to the public in March 2023 running on LaMDA and later PaLM 2. After a widely reported delay, Google DeepMind unveiled Gemini on December 6, 2023 as the company's first natively multimodal model family, initially shipped in three sizes — Ultra, Pro, and Nano — with developer access via the Gemini API and Google AI Studio following a week later. A fine-tuned Gemini Pro immediately began powering Bard.

On February 8, 2024, Google formally retired the Bard brand, renaming the assistant Gemini and launching the Gemini Advanced subscription tier giving access to the larger Ultra 1.0 model. That same month brought the product's first major controversy: a newly launched image-generation feature produced historically inaccurate and racially skewed depictions of real historical figures, such as showing non-White people in Nazi-era uniforms. The backlash — amplified by figures like Elon Musk accusing the model of 'anti-civilizational programming' — forced Google to pause people-generation entirely while it reworked the system, and reporting tied the episode to a roughly $90 billion single-day hit to Alphabet's market value. Around the same period, Gemini 1.5 introduced a mixture-of-experts architecture and a breakthrough long-context window.

Through 2024 into 2025, Google pushed Gemini toward 'reasoning' and agentic capabilities. Gemini 2.0 Flash appeared experimentally in December 2024, and Gemini 2.5 Pro launched as an experimental preview on March 25, 2025, introducing 'thinking' models that reason before answering and debuting at #1 on the LMArena leaderboard. This era also saw Google's image and video generation lines mature — Nano Banana (built on Gemini's image models) and the Veo video series — while Project Astra pushed real-time, camera-based multimodal assistance.

Growth brought new scrutiny. In November 2025, a proposed class action (Thele v. Google) alleged Google secretly enabled Gemini 'smart features' by default across Gmail, Chat, and Meet in October 2025, letting the AI analyze private communications without clear consent; a federal judge later ruled against the plaintiffs for failing to show concrete harm. Separately, conservative activist Robby Starbuck sued Google for defamation, alleging Gemini and related Google AI tools repeatedly generated false claims linking him to child-abuse allegations. Despite the controversies, the Gemini app's user base kept climbing, reportedly surpassing 950 million monthly users by mid-2026.

Google 3.x models arrived in a rapid cadence: Gemini 3 Pro launched November 18, 2025, followed by Gemini 3 Flash in December 2025, then a fast sequence of Flash updates (3.1, 3.5, 3.6, and 3.7) through mid-2026. This acceleration coincided with upheaval at the top: in August 2026, Google DeepMind CEO Demis Hassabis stepped back from day-to-day operations to become GDM chair and Alphabet's chief scientist, handing Gemini development to CTO Koray Kavukcuoglu, while longtime chief scientist Jeff Dean departed to found a rival AI research venture — moves some observers read as a sign of internal strain even as Google touted 'great progress' on an unreleased Gemini 4. Meanwhile the promised flagship Gemini 3.5 Pro slipped past its original mid-2026 target, leaving the fast, cheap Gemini 3.7 Flash — released August 13, 2026 — as the newest and most capable model actually shipping, even as commentators noted Google was still 'trail[ing] Anthropic and OpenAI at the frontier.'