While everyone waits for Gemini 3.5 Pro, Google just shipped three models nobody asked for and quietly registered a fourth that might actually matter.
The Summary
- Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber while its flagship 3.5 Pro remains stuck in testing limbo
- Flash Cyber delivers 42% better security benchmark performance but only works with Google's own API, not through third-party providers like OpenRouter
- The real story: Google internally registered model IDs for both 3.6 Flash and 3.5 Flash-Lite days before the public release, suggesting a deliberate strategy to flood the inference market while competitors wait on flagship models
- Lower costs and faster speeds position these as agent infrastructure plays, not consumer AI plays
The Signal
Google isn't delaying its Pro model because it can't ship. It's shipping everything else because it can't wait. The company released three Flash variants in a single day, each targeting a different price point and use case. This is the AI equivalent of carpet bombing: saturate every inference tier, lock in developers building agent workflows, and worry about the flagship later.
The move makes sense when you map it to where AI spend is actually going. Consumer chatbots aren't the growth vector anymore. The money flows to companies building agents that run thousands of inference calls per day, need predictable costs, and can't afford to wait three months for a Pro model that might get delayed again. Flash-Lite and 3.6 Flash both push speeds up and costs down, which matters more to agent builders than benchmark leaderboard positions.
"Google ships everything but the flagship, and somehow it's the right call."
But here's the tell: Flash Cyber only runs through Google's API, not aggregators like OpenRouter. That's a lock-in play disguised as a security feature. If you're building smart contract auditing tools or on-chain fraud detection, you now have a model fine-tuned for exactly that, but you have to run it through Google's infrastructure. No mixing providers. No cost shopping across platforms. You get the specialized model, Google gets your API call volume and the data exhaust that comes with it.
The Flash Cyber specs tell you what Google thinks is next: 42% better performance on security benchmarks means they see cybersecurity and smart contract auditing as high-margin verticals worth custom models. Not general reasoning. Not creative writing. Security. The kind of work where accuracy matters more than vibes, and where customers will pay for narrow, domain-specific intelligence instead of general-purpose chatbots.
Then there's the quiet part. Google registered model IDs for both 3.6 Flash and Flash-Lite internally before announcing them. That's not news, that's standard practice. What's notable is the pattern: register, test, ship fast. Meanwhile, 3.5 Pro has been "coming soon" for long enough that developers have stopped waiting. Google is prioritizing time-to-market for agent infrastructure over performance benchmarks for Pro users. That's a structural bet on where the revenue is.
Key implications for the agent economy:
- Specialized models (like Flash Cyber) matter more than general-purpose flagships for on-chain work
- API lock-in is back, dressed up as vertical optimization
- The race isn't "who has the smartest model" anymore, it's "who ships inference tools agents can actually use at scale"
The Implication
If you're building agents that need to call models thousands of times a day, Google just handed you three new options optimized for cost and speed. Test them. The performance gap between Flash and Pro might not matter for your use case, and the cost savings probably do. If you're in crypto or security, Flash Cyber is worth the API lock-in for now, but start planning your exit before Google decides to reprice it.
For everyone else: watch what Google ships, not what it promises. The company just told you it thinks agent infrastructure and vertical security models are more valuable than winning the benchmark leaderboard. That's where the money is moving. The Pro model will come when it comes. The market isn't waiting.