Gemini 3.5 Flash drives Google’s AI push against rivals

Google unveiled Gemini 3.5 Flash and a fresh crop of AI agents at I/O on Tuesday, making a broader argument about market power: the next phase of generative AI may hinge less on who has the flashiest chatbot than on who can place fast models inside the products people already use at scale. For Google, that means Search, Android, Workspace and a growing Gemini subscription business.

The company used a faster model, agent tools and deeper Search integration to show investors, developers and enterprise buyers that Gemini is becoming a distribution layer — not a gadget parade but a platform push. OpenAI and Anthropic still shape much of the conversation in frontier AI, but neither starts with Google’s installed base in search queries, browsers, phones and productivity software.

Publishers and other skeptics read the same launch differently. If AI moves from a chat tab into the search box itself, the gain for Google can be the loss of everyone who depends on outbound clicks. The real tension in the I/O announcements sits between Google’s scale advantage and the wider web’s traffic model, a split visible in TechCrunch’s assessment of the new Search experience and The Verge’s rundown of the Search redesign.

Developers and power users face a simpler question. If Flash is fast enough to stay on by default and cheap enough to spread across coding, search and research tasks, habits may change. If the agent layer feels bolted on, then Google will have added one more interface to a crowded market. Whether Tuesday’s launch was a product cycle or a platform turn will depend on that question, not the size of the keynote applause.

Distribution, not just models

Google’s official Gemini 3.5 launch post presented Flash as a model built for coding and “agentic” work, while Reuters reported that the company used the same event to pitch enterprises on lower-cost access and consumers on a broader Gemini app. Speed and price help win developer trials; distribution helps keep those trials inside Google’s own stack.

By its own account, Flash is up to four times faster than other frontier models on some tasks. Google also tied the model to the Gemini API, AI Studio, Android Studio and, according to Engadget’s coverage of the rollout, its Antigravity tooling. The company is not only trying to sell a model. It is trying to reduce the number of handoffs between search, coding, automation and payment so that Gemini becomes the default path through all four.

Pricing drew almost as much attention as model capability. CNBC reported that Google introduced a $200-a-month AI Ultra tier, while Reuters reported that the Gemini app now has 900 million monthly users, AI Overviews reach 2.5 billion monthly users and AI Mode reaches 1 billion monthly users. Those are distribution numbers, not lab numbers. They suggest Google believes it can do to AI what it long did in search and mobile: use reach to spread features quickly, then sort out monetisation across subscriptions, ads and enterprise tools.

In the official launch post, Google made the trade-off explicit before insisting it had narrowed it.

“you no longer have to trade quality for latency.”
— Google

The harder question for analysts is what happens after the speed gains. Developers will tolerate weaker models for lower latency, but only up to a point. Google’s bet is that a model that feels instant, stays cheap and sits inside products users already open each day may be more commercially durable than a stronger model that still requires a separate destination. Whether lower pricing widens Google’s moat or simply compresses margins across the sector is the question at the centre of the I/O event.

Competition was the point. Reuters framed the launch against OpenAI and Anthropic, while CNBC said the company used I/O to show Wall Street that Gemini could move beyond a chatbot. Google is responding to rivals that won mindshare early, especially in coding and general-purpose assistants. The answer it offered on Tuesday was not a single knockout model. It was a bundle: faster inference, agents, subscriptions, developer tooling and tighter product integration.

Search becomes the battleground

Search may be the bigger strategic move. Alphabet said last quarter that Search queries hit an all-time high and revenue rose 19 per cent, giving the company room to push harder on AI inside its most important business. At I/O, Google used that position to argue that AI is not replacing Search so much as remaking it from within.

Pichai’s line, as Reuters reported, was more revealing than the product names.

“When people use our AI-powered features in Search, they use Search more.”
— Sundar Pichai

Should that pattern hold, Google gets something OpenAI and Anthropic cannot easily match: an AI funnel attached to the web’s largest discovery engine. The company can turn Search into a live distribution mechanism for Gemini, a place to train user habits around agents and, eventually, a layer that sends people to fewer websites because more tasks end inside Google’s own interface. TechCrunch’s report on the new information agents and Wired’s account of the agentic search push both pointed to that shift.

A search box that becomes an AI workspace can be efficient for users and punishing for publishers — that is the skeptic camp’s core objection. Google can say overall search usage rises, and the statement may be true, yet the click economy can still weaken if answers, research summaries and task-completion tools keep users on Google’s pages longer. That concern sat behind TechCrunch’s blunt verdict that traditional search is being displaced and The Verge’s view that Search is entering its biggest redesign in years.

The company still has to prove the workflow

The strongest argument Google can make is that consumers and developers do not want isolated AI products. They want faster help inside the tools they already use. That is the logic behind Flash, Spark, Search AI Mode and the broader agent pitch. In its launch material, Google said agents should be capable of “taking action on your behalf while under your direction”. The ambition is clear. The harder part is reliability.

From a developer’s view, the standard is likely to be pragmatic. If Flash reduces waiting time, if Antigravity and coding tools remove setup friction and if the agent products handle routine work without constant supervision, Google will have a strong claim to everyday use. If those features stay closer to demos than workflows, then OpenAI and Anthropic can keep competing on model quality and specialist loyalty even without Google’s scale. The most important comparison after I/O will not be keynote breadth. It will be whether developers and office users actually keep these tools open a week later.

Not every part of the bundle has to win. A fast model can lift Search. Search can feed subscriptions. Subscriptions can finance broader deployment. Developer tools can make the platform stickier. That is a platform strategy, not a gadget launch.

Yet the same concentration that makes the strategy powerful also sharpens the backlash around web traffic, gatekeeping and dependency on one company for search, productivity and AI assistance at once. Tuesday’s announcements showed that Google understands the market is moving from chatbot novelty to workflow control. The question now is whether that control proves useful enough for users, open enough for developers and tolerable enough for the rest of the web.