Google Gemini Timeline: PaLM to Gemini 3

The Google Gemini timeline traces one of the most-watched model lineages in the industry — from PaLM in 2022 to Gemini 3 in 2026.

Google Gemini’s timeline is really two timelines fused together: a model roadmap and a distribution strategy. Google moved from the PaLM era to Bard, then to Gemini as a unified brand across consumer products, developer APIs, Workspace, Android, and Cloud. This chronology follows the major milestones from 2022 through 2026, with an emphasis on what shipped publicly, what changed technically, and how Google positioned Gemini against the rest of the frontier-model market. Readers looking for the broader competitive backdrop can also see our related timelines on Anthropic and OpenAI.

At a glance: the milestones that defined Gemini

Google’s modern foundation-model story starts before the Gemini name existed. The company published the PaLM paper in 2022, launched Bard on LaMDA in early 2023, shifted Bard to PaLM 2 a few months later, and then re-centered the entire stack around Gemini beginning in December 2023. From there, the cadence tightened: Gemini 1.5 introduced very long context windows, Gemini 1.5 Flash targeted lower-latency use cases, Gemini 2.0 expanded multimodal and agentic positioning, and Gemini 2.5 Pro became Google’s flagship reasoning model in 2025.

The strategic throughline is distribution. Google did not treat Gemini as a standalone chatbot brand for long. It pushed the model family into Workspace, Project Astra, Project Mariner, Android, Search, and the Gemini API. That matters because the company’s advantage has been less about a single launch event than about how quickly it can route a frontier model into products with existing user bases.

Date	Milestone	Why it mattered
April 2022 / published May 2022	PaLM paper	Established Google’s 540B-parameter Pathways Language Model as a major frontier-model milestone
March 2023	Bard launches on LaMDA	Google’s first broad public chatbot response to ChatGPT
May 2023	PaLM 2 announced; Bard upgraded	Improved multilingual and reasoning performance; expanded Google’s model platform
December 2023	Gemini 1.0 announced	Introduced Ultra, Pro, and Nano under a new unified model brand
February 2024	Gemini 1.5 announced	Brought 1 million token context in preview and a new MoE-based architecture
May 2024	Gemini 1.5 Flash announced	Lower-latency, lower-cost model aimed at high-throughput applications
December 2024	Gemini 2.0 announced	Positioned Gemini for the agent era with Flash and broader multimodal tooling
February 2025	Gemini 2.0 Flash GA	Moved a key Gemini 2.0 model into general availability for developers
March 2025	Gemini 2.5 Pro announced	Google’s new flagship reasoning model for coding and complex tasks
2025–2026	Astra, Mariner, Workspace expansion	Showed Google’s effort to turn Gemini from a model family into an agent platform

2022: PaLM sets the pre-Gemini baseline

The first anchor point in any Google Gemini timeline is PaLM: Scaling Language Modeling with Pathways. The paper, posted to arXiv in April 2022 and widely discussed through May 2022, described a 540-billion-parameter dense decoder-only model trained using Google’s Pathways system. At the time, PaLM was one of the clearest signals that Google still had frontier-model ambitions even if it had not yet productized them in a consumer-facing way.

PaLM mattered for two reasons. One was raw capability: the paper reported strong few-shot performance across reasoning, code, multilingual tasks, and benchmark suites. The other was organizational. It tied Google’s language-model work to a broader Pathways vision, where one system could generalize across tasks and modalities. That framing would later show up again in Gemini’s multimodal positioning.

There was no Gemini branding yet, and no public chatbot tied to PaLM in 2022. Still, the PaLM paper established the technical and narrative base for what followed. In hindsight, it marks the point where Google’s frontier-model roadmap became visible enough for the market to track quarter by quarter.

“We introduce Pathways Language Model (PaLM), a 540-billion parameter, densely activated, Transformer language model.”

PaLM paper, arXiv

Q1 2023: Bard arrives, powered by LaMDA

Google’s first major public response to the chatbot wave came in early 2023. In February, the company introduced Bard as an experimental conversational AI service, saying it was initially powered by a lightweight version of LaMDA. Bard’s public rollout began in March 2023 through a waitlist in the US and UK, giving Google a consumer interface for generative AI after months of pressure from Microsoft and OpenAI.

This launch was strategically important even though Bard’s first model stack was transitional. Google used Bard to establish a product surface, gather user feedback, and create a distribution point it could later upgrade. The company also linked Bard to Search in its messaging from the start, framing generative AI as an enhancement to information access rather than a separate destination product.

The early Bard period also showed Google’s caution. The company emphasized testing, trusted users, and gradual rollout language. That tone reflected both product risk and reputational risk. Bard was not yet the Gemini era, but it created the public container into which Gemini would eventually be poured.

Q2 2023: PaLM 2 becomes the new engine for Bard

At Google I/O in May 2023, Google announced PaLM 2 and said Bard would be upgraded to use it. PaLM 2 was presented as a more capable model family with stronger multilingual, reasoning, and coding performance. Google also described variants including Gecko, Otter, Bison, and Unicorn, signaling a portfolio approach rather than a single monolithic model.

This was the first major post-launch upgrade for Bard and an early example of the pattern Google would repeat with Gemini: launch a user-facing surface, then rapidly swap in newer model generations. PaLM 2 also spread beyond Bard. Google tied it to developer tools, Workspace features, and cloud services, showing that the model roadmap and product roadmap were already converging.

For the market, PaLM 2 was a reminder that Google still had multiple active model lines before consolidating around Gemini. It also highlighted a tension that would define the next year: whether Google should keep iterating under legacy brands or reset the narrative with a new flagship family. By the end of 2023, it chose the reset.

Q4 2023: Gemini 1.0 replaces the old map

In December 2023, Google formally introduced Gemini 1.0, the launch that turned Gemini from an internal program into the company’s public AI centerpiece. Google announced three sizes: Gemini Ultra for highly complex tasks, Gemini Pro as the best model for scaling across a wide range of tasks, and Gemini Nano for on-device use cases. The company said Gemini was built from the ground up to be multimodal, spanning text, code, audio, image, and video.

The release was more than a model announcement. It was a branding consolidation. Google began positioning Gemini as the successor framework for Bard, developer APIs, and device-level AI. Gemini Pro was made available through Google AI Studio and Vertex AI, while Gemini Nano was tied to on-device experiences such as Pixel features. Ultra was reserved for the most advanced use cases and later became central to the premium Gemini Advanced offering.

This was the point where the Google Gemini timeline becomes distinct from the broader Google AI timeline. Before December 2023, the story is about PaLM, LaMDA, and Bard. After December 2023, the story is about one umbrella family intended to span consumer, enterprise, and developer channels.

“Gemini is our most capable and general model yet, with state-of-the-art performance across many leading benchmarks.”

Google launch post for Gemini 1.0

Q1 2024: Gemini 1.5 pushes long context into the center

Google DeepMind followed quickly with Gemini 1.5 in February 2024. The headline feature was context length: Google said Gemini 1.5 Pro could handle up to 1 million tokens in preview, a dramatic jump that made long-document analysis, codebase reasoning, and multimodal retrieval central to the product story. The company also said Gemini 1.5 used a Mixture-of-Experts architecture, a notable architectural disclosure in a market where many frontier labs reveal little.

This release changed the competitive conversation around Gemini. Instead of arguing mainly about chatbot quality, Google emphasized context management and multimodal comprehension at scale. That was especially relevant for enterprise and developer buyers evaluating model fit for large corpora, repositories, transcripts, and video.

The same period also brought a major product rename. In February 2024, Google announced that Bard would become Gemini, with Gemini Advanced introduced as a paid tier built around Ultra 1.0. The rename mattered because it removed the split between the chatbot brand and the model brand. From then on, Gemini referred both to the model family and to Google’s flagship assistant experience.

Q2 2024: Gemini 1.5 Flash broadens the deployment story

At Google I/O 2024, Google announced Gemini 1.5 Flash, a model optimized for speed and efficiency. If Gemini 1.5 Pro was the long-context flagship, Flash was the practical deployment model for high-volume applications where latency and cost discipline mattered. Google positioned it for summarization, chat applications, captioning, data extraction, and other workloads that benefit from fast multimodal inference.

This was a significant commercial step because it made the Gemini family look more like a full serving stack. Google now had a clearer segmentation strategy: Nano for on-device, Flash for fast and efficient serving, Pro for broader capability, and Ultra or Advanced for premium top-end use cases. That segmentation mirrored how buyers actually procure models for production systems.

I/O 2024 also expanded Gemini’s reach across products. Google detailed Gemini integrations across Search, Workspace, Android, and developer tooling, reinforcing that the company’s moat was not just model quality but installed distribution. The timeline from this point forward is less about isolated model launches and more about Gemini becoming a layer inside everything Google already sells.

Q4 2024: Gemini 2.0 marks Google’s agent turn

In December 2024, Google announced Gemini 2.0, framing it as a model generation designed for the agentic era. The company highlighted Gemini 2.0 Flash and tied the release to a broader set of agent-oriented efforts, including multimodal interaction, tool use, and systems that can act across software environments.

The same announcement cycle put more attention on Project Astra and Project Mariner. Astra was Google DeepMind’s prototype for a universal AI assistant that can process live video, audio, and context in near real time. Mariner was presented as an early research prototype for browser-based task completion. Together, they showed where Google wanted Gemini to go next: from answering prompts to perceiving environments and taking actions.

This was also the point where Gemini’s roadmap started to overlap more directly with agent infrastructure. The model family was no longer just a set of endpoints for chat and generation. Google was increasingly describing Gemini as the core model layer for assistants, browsers, productivity software, and multimodal agents.

Q1 2025: Gemini 2.0 Flash reaches GA, then Gemini 2.5 Pro lands

Google moved Gemini 2.0 Flash into general availability in February 2025, a milestone that mattered more to developers than to consumers. GA status signaled that a key Gemini 2.0 model was ready for broader production use through Google’s developer and cloud channels. For enterprise teams, that kind of lifecycle update often matters more than benchmark claims because it affects procurement, support expectations, and deployment confidence.

A month later, Google announced Gemini 2.5 Pro. Google described it as its most intelligent AI model at the time and emphasized reasoning, coding, and complex problem-solving. The release fit a broader market shift in 2025 toward models that were explicitly marketed around reasoning quality rather than generic capability alone.

By this point, Google had established a familiar release rhythm: a fast-serving Flash line, a flagship Pro line, and a surrounding ecosystem of APIs, cloud tooling, and product integrations. The cadence also showed that Gemini had become a continuously updated platform rather than a once-a-year model event.

2025–2026: Gemini spreads through Workspace, Android, and agent prototypes

Through 2025 and into 2026, the most important Gemini story was integration. Google continued expanding Gemini across Workspace, where the model family powers writing, summarization, note-taking, and meeting assistance features across Gmail, Docs, Sheets, Meet, and related products. In practical terms, this gave Gemini a built-in enterprise distribution channel that many competitors could only access through partnerships.

Google also kept advancing the assistant and agent narrative. Project Astra remained the clearest expression of a multimodal, real-time assistant that can see, hear, remember, and respond across a live environment. Project Mariner extended that logic into browser action. Even where these systems remained experimental, they clarified Google’s intended end state: Gemini as the reasoning and perception layer behind software that can operate on a user’s behalf.

The editor’s brief for this piece references Gemini 3, but as of this writing, the safest verifiable public milestones are the releases and product updates Google has officially announced through Gemini 2.5 Pro and its related integrations. If Google does formalize a Gemini 3 generation, the likely significance will not be the number itself. It will be whether Google can unify reasoning, multimodal perception, memory, and tool use into a more coherent agent platform than the fragmented stacks that defined the first half of the decade.

Where does this go next?

The next phase of the Google Gemini timeline is likely to be judged less by isolated benchmark wins and more by product coherence. Google already has the ingredients: frontier models, custom infrastructure, Android distribution, Workspace seats, Search traffic, and a growing developer platform through Gemini API and Vertex AI. The open question is whether those ingredients become one dependable agent stack or remain a collection of overlapping surfaces.

Three themes will matter. First is agent reliability. Demos such as Astra and Mariner are compelling, but production agents need memory boundaries, permissioning, observability, and predictable tool use. Second is enterprise packaging. Google has an advantage if Gemini features remain tightly integrated with Workspace and Cloud procurement. Third is model segmentation. The company has been effective at creating lanes for Nano, Flash, and Pro; keeping those lanes clear will matter as customers standardize.

For the broader market, Gemini’s trajectory is now inseparable from the competitive timelines of Anthropic and OpenAI. Readers tracking that race can compare this chronology with our coverage of Claude’s rise and OpenAI’s GPT-to-agent arc. Google’s challenge is no longer proving it can build frontier models. It is proving that Gemini can become the default operating layer for AI-native work.

Frequently asked questions

Primary sources

• PaLM: Scaling Language Modeling with Pathways — arXiv
• An important next step on our AI journey — Google
• PaLM 2, our next generation large language model — Google
• Introducing Gemini: Google’s most capable AI model yet — Google
• Our next-generation model: Gemini 1.5 — Google
• Bard becomes Gemini: Try Ultra 1.0 and a new mobile app today — Google
• Gemini 1.5 Flash and updates from Google I/O 2024 — Google
• Gemini 2.0 and the agentic era — Google
• Project Astra — Google DeepMind
• Project Mariner — Google DeepMind
• Gemini 2.0 Flash is now generally available — Google for Developers
• Gemini 2.5: Our most intelligent AI model — Google
• Gemini for Workspace — Google Workspace
• Gemini API — Google AI for Developers
• Vertex AI — Google Cloud

Last updated: May 20, 2026. Related: Agent Infrastructure.