Google Just Became Your Second Vendor

For a year, planning around AI meant planning around OpenAI, with Anthropic as your backup. Google was the thing you kept meaning to test. That math changed this quarter. Gemini is closing on a billion users, the new Flash models are cheaper and stronger, and Google shipped a security agent that undercuts the frontier labs on price. Let me catch you up on what that means for your roadmap.

The scale that makes it safe to bet on

Gemini crossed 950 million monthly users, up from 750 million in February. Google shared the number on its Q2 2026 earnings call, and it puts Gemini right behind ChatGPT, which hit a billion in June. Usage tripled in a year.

The market share numbers back it up. Sensor Tower's State of AI report found ChatGPT's share among AI assistants dropped below 50% for the first time, while Gemini climbed to 27.7%. On iOS alone, the Gemini app pulled 137 million downloads in the last 12 months.

The point for you is not the horse race. It is that a vendor at this scale is not going anywhere. You can design around Gemini without betting on a startup that might get bought or run out of runway. That is what makes a second option real.

Cheap stopped meaning weak

The old rule was simple. The cheap tier, the Flash or Mini or Haiku, traded capability for cost. You called it when you could not afford the smart model. Gemini 3.6 Flash broke that rule on the one benchmark where you would not expect it to.

On OSWorld-Verified, which tests whether an agent can actually drive a real desktop and finish a multi-step task, 3.6 Flash scored 83%. That beat GPT-5.6 and Grok 4.5. A model that costs $7.50 per million output tokens posted the highest computer-use score, not the highest cheap score.

Google also cut the price below the last Flash and made it use fewer tokens. Per its own launch numbers, 3.6 Flash uses 17% fewer output tokens than 3.5 Flash, and up to 65% fewer on some coding benchmarks. Cheaper and better in the same release changes what you can afford to ship at scale.

The flagship that did not show up

Here is the gap in the story. Google shipped three new Flash models and skipped the one you might have wanted. There is no Gemini 3.5 Pro, the high-capability model for complex reasoning, last updated back in February.

Google teased Pro for June. Bloomberg reported the team hit internal delays trying to meet its own performance goals. Product lead Logan Kilpatrick says Pro is testing with partners and hopes to land soon. Meanwhile OpenAI pushed out GPT-5.5 and GPT-5.6, and Anthropic shipped Claude Opus 4.8, Sonnet 5, and Fable 5.

So read the signal plainly. Google is strong right now on high-volume, cost-sensitive agent work. If your roadmap leans on top-end reasoning, Gemini is not your lead vendor this quarter. Match the tool to the job instead of picking one flag to plant.

The security agent that undercuts the frontier

Google also shipped 3.5 Flash Cyber, a model tuned to find and patch code vulnerabilities. Google pitches it as a cheaper alternative to Anthropic's Mythos, which costs twice as much as Claude Opus 4.8 to run.

The results are concrete. Inside Google's CodeMender agent, 3.5 Flash Cyber found 55 confirmed issues in the V8 JavaScript engine, against 47 for regular 3.5 Flash and 36 for Opus 4.6. It found 10 that no other model caught. The trick is calling a cheap model many times instead of one expensive model once.

The catch: it is locked to governments and trusted partners in a limited pilot, so you cannot buy it yet. But it shows Google's play. Undercut the frontier labs on price by leaning on speed and volume.

The agent got 90% cheaper to try

Gemini Spark, Google's agent that reads your email, calendar, and photos to act for you, dropped from the $100 to $200 Ultra plan to the $20 AI Pro tier in the US. That is a price cut of as much as 90%, and it puts Spark at parity with ChatGPT Work and Claude Cowork, both around $20.

One real gap: those rivals work in Europe and the UK. Spark does not reach the EEA, UK, Switzerland, or Nigeria yet. If your team or users sit there, that decides it for you.

The deep cut

The move that changes your Monday is not switching vendors. It is splitting your workload by job. Google just proved a cheap model called many times can beat an expensive model called once, on desktop tasks and on security scans. That means your high-volume agent traffic, the calls you make ten thousand times a day, no longer has to run on your priciest model. Route that to 3.6 Flash or 3.5 Flash-Lite, keep your top reasoning work on whatever leads today, and stop paying flagship prices for grunt work. The savings are real and the quality gap closed.

Three questions for your team

1. Which of our agent calls are high-volume and cost-sensitive, and what would we save moving them to 3.6 Flash or 3.5 Flash-Lite this quarter?

2. Are we designing our stack around one vendor by habit, or can we route each job to the model that wins it, given Google is now a safe bet at scale?

3. If our users or team sit in Europe or the UK, does Spark's missing coverage rule it out, and does that change which agent we build against?